Gemini API Credits for Startups: Every Route in 2026

Where a startup can get its Gemini API bill paid in 2026, which credits actually reach the Gemini API, and the billing rules that decide whether a grant works at all.

Google's $300 free trial stopped paying for the Gemini API on March 2, 2026. Google's billing documentation is direct about it: "If you opened your account after March 2, 2026, you can't use these credits to pay for Gemini API and AI Studio usage" - Gemini API billing docs. The $300 still arrives with every new Google Cloud account, and it still pays for plenty, including Gemini through Google's enterprise AI platform. It just no longer pays for the API most developers start with, and founders keep finding out from a card statement. One founder on Google's developer forum reported in March that their Gemini API charges "were NOT covered" by trial credits, "resulting in a real charge to my credit card" - Google AI Developers Forum.

The second surprise is bigger, because it hits funded startups. Google's startup program does pay for the Gemini API, and says so: its Gemini page invites founders to apply for "Cloud credits (from $2k to potentially $350k), to use towards your Gemini API usage as you scale" - Google for Startups. But since Google moved the Gemini API to prepaid billing, those credits only flow while you also hold cash in a prepaid balance: "When your Prepay credit balance on the billing account hits $0, Google Cloud credits will no longer be consumed" - Gemini API billing docs. A startup sitting on $350,000 of credits and a $0 prepaid balance gets an HTTP 402 error, not a working API.

This guide maps every route we verified on its owner's own page on October 12, 2026: the free tier and what it costs you in data, the $300 trial and the one door it still opens, the Google for Startups Cloud Program's Start, Scale and AI tiers, accelerators and the AI Futures Fund, hackathons, research programs, gateways like OpenRouter and Vercel, and banking perks. It explains the 2026 billing system (Prepay, usage tiers and the $250 monthly cap on new accounts) that sits between a grant and a working API, prices the credits in real requests on the current models, and ends with a decision framework, the order to apply in and the mistakes that waste credits. Every number links to the page it came from.

Contents

1. The short answer: every route, scored
2. How Gemini API billing works in 2026: free tier, Prepay, tiers and caps
3. Route 1: the free tier and AI Studio
4. Route 2: the $300 Google Cloud trial, and the one door it still opens
5. Route 3: the Google for Startups Cloud Program
6. Route 4: accelerators, the AI Futures Fund and investor packages
7. Route 5: hackathons, sprints and challenges
8. Route 6: research, education and nonprofit programs
9. Route 7: gateways, platforms and banking perks
10. What Gemini credits buy: the burn math
11. Making credits last: model choice, Batch, Flex and caching
12. Spend caps: protecting the grant from a runaway bill
13. Which route fits your startup
14. Stacking routes, and the order to apply
15. Eight mistakes that waste Gemini credits
16. Frequently asked questions
17. How we verified this guide

1. The short answer: every route, scored

Gemini is unusual among frontier models in one structural way: Google sells it, and Google also runs the largest startup credit program that can pay for it. With Claude or GPT, a founder has to find a cloud or gateway willing to let its credits cover another company's model. With Gemini, most of the money that can reach the model is Google's own, so the real questions are narrower. Which of Google's credits does Google let you spend on the Gemini API, through which of its two doors, and under which billing rules?

The table below ranks every route we verified on four criteria. Gemini fit (30%) asks whether the money pays for the Gemini API directly, through Google's enterprise platform only, or through a reseller. Amount (25%) is the published value. Access (30%) is how many startups can actually get it today, including whether applications are open. Flexibility (15%) covers duration, expiry and conditions. Each cell carries the fact behind its score, and the final column is the weighted average.

Read the table as a sequence, not a menu. The free tier tops it because every startup can use it today at no cost, which says more about how hard the paid routes are to reach than about how much the free tier gives. The routes that put real money behind production traffic are rows 2 and 3, Google's own program, and both run through the same Prepay mechanics described in section 2. Everything below row 5 is either small, closed, or limited to researchers.

The chart is a picture of concentration. Google's own program is more than twenty times larger than anything else that can pay for Gemini, and the next-largest route, Vercel's, is a platform credit of which at most half can go to model calls. The "OpenRouter perk" bar is a separate offer inside Google's own program: Scale-tier members can receive "$1k in OpenRouter credits for use on Google models, valid for 6 months" - Google Cloud Startup Perks. The trial's $300 bar comes with an asterisk the chart cannot show: it no longer touches the Gemini API itself.

Why the map looks like this comes down to who bills the tokens, the same principle that decides every startup credit question. A credit is a customer-acquisition subsidy, and a company subsidizes spend it collects. Google collects every dollar of Gemini spend, whichever of its two doors you use, so all of Google's credits could in principle pay for Gemini. In practice Google has drawn lines inside its own house: the trial credit stops at the Developer API, startup credits wait for a prepaid balance, and new billing accounts are capped. Those lines are product decisions about which customers Google wants to subsidize, and once you read them that way, the rules stop looking arbitrary. The trial targets people evaluating Google Cloud as a whole; the startup program targets companies Google expects to become large customers; and the Prepay floor makes sure every account that burns credits has a real payment relationship behind it.

How to apply this section: start on the free tier today, apply for the Google for Startups tier you qualify for as soon as you have a product, and treat everything else as top-ups. If you are funded and AI-first, the Scale AI tier is worth more than every other route in this guide combined, and the rest of this guide is mostly about making sure that grant actually reaches your API calls.

2. How Gemini API billing works in 2026: free tier, Prepay, tiers and caps

Every route in this guide ends at the same meter, so it pays to understand the meter before chasing the money. The Gemini Developer API, the API you reach with a key from Google AI Studio, runs on Google Cloud's billing system: "The Gemini API uses Cloud Billing accounts for billing services, which you can set up directly in AI Studio" - Gemini API billing docs. A project with no billing account sits on the free tier; a project linked to a billing account sits on the paid tier. Everything else, including how startup credits are applied, follows from which tier a project is on and how its billing account pays.

Google rebuilt that billing account in 2026, and most guides online still describe the old one. Three changes matter. Paid usage is now prepaid by default. Spending is now capped per billing account according to a usage tier. And the Google Cloud $300 Welcome credit no longer pays for the Gemini API at all on accounts opened after March 2, 2026. Each change moves money in a way a founder holding credits will feel, so this section takes them one at a time.

Free tier versus paid tier: the price is your data

The free tier is real, and for prototyping it is generous. Google's pricing page lists it as "Limited access to certain models", "Free input & output tokens" and "Google AI Studio access" - Gemini API pricing. The fourth line on the same card is the cost: "Content used to improve our products". Every model table on that page carries a row titled "Used to improve our products", answered "Yes" for the free tier and "No" for the paid tier.

The terms of service spell out what that means. "When you use Unpaid Services, including, for example, Google AI Studio and the unpaid quota on Gemini API, Google uses the content you submit to the Services and any generated responses to provide, improve, and develop Google products", and "human reviewers may read, annotate, and process your API input and output" - Gemini API Additional Terms. The same terms tell you plainly: "Do not submit sensitive, confidential, or personal information to the Unpaid Services." The paid tier flips this: Google "doesn't use your prompts... or responses to improve our products", and logs them only for a limited period to detect abuse.

Two consequences follow for a startup. The first is that the free tier is a development tool, not a production plan for anything that touches customer data. The second is regional. The terms add: "You may use only Paid Services when making API Clients available to users in the European Economic Area, Switzerland, or the United Kingdom." You can build on the free tier from Berlin or London, as Google's billing FAQ confirms, but the moment your product serves users there, it has to run on the paid tier.

Prepay: credits you buy before you call

The second change is Prepay. Google's billing docs say "New users default to the Prepay billing plan", that upgrading from the free tier "means linking a billing account and prepaying to add a minimum of $5" of credits, and that the most you can hold is $5,000 - Gemini API billing docs. Purchased credits "expire after 12 months and are non-refundable", and they "apply only to Gemini API usage costs; you can't use them to pay for other Google Cloud services."

The failure mode is abrupt. "When your Prepay credit balance on the billing account hits $0, all API keys in all projects linked to that billing account will stop working simultaneously", and requests "fail with an HTTP 402 Payment Required error until you add credits." Google offers an auto-reload that tops the balance up when it runs low, and a monthly auto-charge limit so a reload loop cannot drain a card. For a startup in production, auto-reload is not optional: a dashboard alert that arrives after the balance hits zero arrives after your product went down.

Prepay is also the rule that decides whether a startup grant works on the Gemini API, which is the single most important detail in this guide. We return to it in section 5, because it changes how you should set up billing before your credits land.

Usage tiers and the billing tier cap

The third change is a spending ceiling per tier. The Gemini API sorts billing accounts into a Free tier and three paid tiers, and each paid tier carries a "billing tier cap" - Gemini API rate limits. Tier 1 requires only that you "Set up and link an active billing account" and is capped at $250. Tier 2 requires "Paid $100 + 3 days from first successful payment" and is capped at $2,000. Tier 3 requires "Paid $1,000 + 30 days from first successful payment" and is capped at "$20,000 - $100,000+".

The billing docs explain what happens at the cap: service "is paused for all projects linked to that billing account until the start of the next billing cycle" - Gemini API billing docs. On top of the monthly cap sits a spend rate limit evaluated on a rolling 10-minute window: $10 for Tier 1, $50 for Tier 2 and $200 for Tier 3, after which the API returns "429 RESOURCE_EXHAUSTED". Tier qualification counts "total cumulative spending on Google Cloud services (including, but not limited to, the Gemini API) for the billing account linked to your project", and Google warns that "in rare cases an upgrade request may be denied based on other factors."

Here is why this matters to anyone holding credits. Tiers belong to the billing account, not to the grant: "Tiers, rate limits, and billing account caps are all determined at the billing account level." A startup that receives a large credit on a brand-new billing account starts at Tier 1, where it can spend at most $250 a month on the Gemini API however large its balance is. Google's pages do not say whether credit-funded spend counts as "paid" toward Tier 2 and Tier 3, so the safe assumption is that a few real payments are what move you up. Per-model rate limits (requests and tokens per minute) are not published on the docs page any more; Google says they "can be viewed in Google AI Studio."

How to apply this section: decide which billing account your Gemini projects will live on before you apply for anything, because tier, cap and history all attach to it. If your company already pays Google Cloud bills, use that account. Turn on auto-reload with a sensible monthly limit, and check your tier on the AI Studio billing page before you plan a launch: a $250 cap is a production outage waiting for its first busy day.

Two doors to the same models: the Developer API and Agent Platform

Gemini is sold through two separate products, and credits behave differently at each. The Gemini Developer API (AI Studio keys, generativelanguage.googleapis.com) is the one this guide has described so far. The other is Google Cloud's enterprise AI platform, which Google renamed in 2026: the old Vertex AI pricing address now redirects to "Gemini Enterprise Agent Platform" - Google Cloud, and Google's documentation now refers to "Gemini Enterprise Agent Platform (formerly Vertex AI)". Older guides that talk about "Vertex AI credits" are describing this second door.

The prices are the same at both doors on the global endpoint. Agent Platform lists Gemini 3.8 Flash at an introductory "$0.75 / $3.75 per 1M tokens input / output through December 31, 2026" and "$1.5 / $7.5" from January 1, 2027 - Google Cloud pricing, matching the Developer API; its non-global regional endpoints cost 10% more. What differs is everything around the price. Agent Platform authenticates with Google Cloud service accounts, bills through ordinary postpaid Cloud Billing rather than Gemini Prepay, and is where Google's enterprise features live. Google's own comparison page is blunt about the default: "Most developers should use the Gemini Developer API unless there is a need for specific enterprise controls" - Google AI for Developers.

For credits, the second door matters more than that sentence suggests. Several credits in this guide pay for Gemini on Agent Platform but not on the Developer API, starting with the $300 trial in section 4. A startup that would otherwise default to an AI Studio key should at least know the other door exists, because switching is a matter of authentication and an endpoint, not a rewrite of the product.

3. Route 1: the free tier and AI Studio

The cheapest Gemini credit is the one you never apply for. A new Google account can create an API key in AI Studio and start calling the API on the free tier with no billing account at all, and Google states that "Google AI Studio usage is free of charge in all available regions" - Gemini API pricing. For a team still deciding whether Gemini fits its product, this is where the evaluation should happen, because nothing in the rest of this guide arrives quickly enough to fund a first prototype.

What the free tier covers changed in 2026, and the change cuts against the way most founders try models: you cannot start on a free Pro model any more. The pricing page marks the current Pro-class model, Gemini 3.1 Pro Preview, as "Not available" on the free tier, alongside every Nano Banana image model and the Gemini Omni Flash video model. The free rows are the Flash family: Gemini 3.8 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, Gemini 3.1 Flash-Lite, plus the Live, text-to-speech, transcription and embedding models. The older 2.5 models still show a free tier on the pricing page, but Google's model page says it is "limiting access to the 2.5 models to users who have actively used them in the past" and tells new projects to "use our latest models: 3.5 Flash-Lite or 3.8 Flash" - Gemini models.

What the free tier costs you

The price of the free tier is your prompts, as section 2 described: Google uses free-tier content to improve its products, and human reviewers may read it. That is a fair trade for experiments and a bad one for a business, because the data a startup's product handles is usually the data it has promised to protect.

The limits are the other cost. Google no longer publishes the free tier's per-model request limits on its documentation pages; the rate-limit page says only that limits "can be viewed in Google AI Studio" - Gemini API rate limits, and that "Rate limits are more restricted for experimental and preview models." Free limits are the lowest Google offers, and they apply per project rather than per key, so spinning up more keys does not buy more capacity. Together, those two costs rule out these kinds of traffic on the free tier, no matter how small the volume:

• Customer data, including support tickets, documents and anything personal
• Proprietary material, such as your own code, contracts or internal knowledge
• Products for EEA, UK or Swiss users, which the terms reserve for paid services
• Production traffic that has to stay up, because free limits are the lowest
• Search-grounded features, since grounding on Gemini 3 is not available on the free tier

The list is short because the rule is simple: the free tier is for your own experiments, not for anyone else's data. Synthetic data, public documents and test prompts are fine. The grounding limit surprises people building research or news features: the free tier shows "Not available" for Grounding with Google Search on the Gemini 3 models, with a footnote that it "Can be tested in Google AI Studio", while the paid tier includes "5,000 free search requests per month (shared across all Gemini 3 and newer models), then $14 per 1,000 requests" - Gemini API pricing.

The free extras around the API

Three adjacent offers stretch the free tier further. The first is deployment: AI Studio's Build mode can publish an app, and Google's Starter Tier "Lets eligible accounts publish up to 2 full-stack applications without setting up a Google Cloud project or billing account", with each deployment creating a Cloud Run service - Google AI for Developers. Workspace accounts and anyone with "an active or prior paid Google Cloud billing account" are not eligible, so this is a perk for brand-new builders rather than established companies.

The second is Firebase AI Logic, Google's client SDK for calling Gemini from web and mobile apps. Firebase says "Using Firebase AI Logic is free-of-charge", and its free path runs on the Gemini Developer API's free tier: "To use this 'free tier', you do not link your project to a Cloud Billing account. This means your Firebase project is on the no-cost Spark pricing plan" - Firebase. Moving to paid limits means the Blaze plan and, in Firebase's words, "Prepayment usually required."

The third is express mode on Google's enterprise platform, the free option fewest founders know about. "New users to Google Cloud can sign up for an express mode account in a free tier to try Agent Platform for free for up to 90 days, within the specified quotas. You don't need to provide billing information" - Google Cloud documentation. It is in Preview, it requires a Gmail account, and it is "separate from, and not available through, the Google Cloud Free Program." Its model list includes Gemini 3.1 Pro Preview, which makes it the one free way we found to test Google's current Pro model, though Gemini 3.8 Flash was not on the list when we read it.

What is no longer free

One popular free route closed this year. Gemini CLI and the Gemini Code Assist IDE extensions used to offer generous free quotas to anyone with a personal Google account, and many founders treated them as free coding agents. Google's developer blog announced that "On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra, as well as those using it free of charge using Gemini Code Assist for individuals", with the CLI remaining usable "via paid Gemini and Gemini Enterprise Agent Platform API keys" - Google Developers Blog. The CLI's README on GitHub still advertises the old free quota, so treat any guide that quotes it as out of date. The successor, Antigravity, describes its free allowance only as "A meaningful quota, refreshed weekly" - Antigravity.

How to apply this section: build your prototype on the free tier with synthetic or public data, measure your real token counts per request while you do it (the burn math in section 10 needs them), and switch to the paid tier the day real users or real data arrive. If you need to evaluate the Pro model before paying, express mode is the way in. Our free LLM API comparison puts Gemini's free tier beside every other provider's, and our Gemini API free tier page tracks its current terms.

4. Route 2: the $300 Google Cloud trial, and the one door it still opens

Every new Google Cloud customer still gets $300 of credit. "The Google Cloud Free Trial is a 90-day program that lets new users try the most popular Google Cloud products", and the trial billing account comes "preloaded with $300 in free Welcome credit which is valid for 90 days" - Google Cloud documentation. You qualify if you have never been a paying customer of Google Cloud, Google Maps Platform or Firebase and have not had the trial before, and Google asks for a card or other payment method up front.

For two years, that $300 was the default way to put a few thousand Gemini calls on someone else's bill, and countless tutorials still say so. It is no longer true for the Gemini Developer API. The trial documentation now lists the exclusion in plain words: "The $300 credit can't pay for Gemini API in AI Studio costs. Consider using the Gemini API Free tier, or sign up for a Prepay billing plan." The Gemini billing FAQ dates it: "starting March 2026, Gemini API usage costs are specifically excluded from the $300 Google Cloud Free Trial program" - Gemini API billing docs.

Who still has trial credits that work

There is a grandfather clause. "If you were granted a Google Cloud welcome credit before they became ineligible, you are allowed to spend your remaining credits on the Gemini API and AI Studio up until the credits expire (after 90 days)." Since the trial lasts 90 days and the cutoff was March 2, 2026, every grandfathered balance has now expired. In practice, as of October 2026, no trial credit pays for the Gemini Developer API.

The cutover produced exactly the confusion you would expect. Forum threads from the months that followed show founders surprised by card charges: one reported a €60.70 charge with no trial credit applied and wrote that "The policy change (around March 7, 2026) was not clearly communicated" - Google AI Developers Forum. Another, in September, watched Gemini usage draw down a $25 prepaid balance while their $300 trial credit "remains untouched" - Google AI Developers Forum. Both are what the documentation now describes, not billing errors.

The door it still opens: Gemini on Agent Platform

The exclusion is narrower than it first looks. It names the Gemini Developer API and AI Studio, not Gemini as a model. Firebase's pricing page, which documents both of Google's Gemini APIs side by side, draws the line exactly: the trial "credit cannot be used towards Gemini Developer API costs, but the credit can be used for the Agent Platform Gemini API (formerly Vertex AI)" - Firebase. The trial terms also exclude "a generative AI partner model that is offered as a managed API" - Google Cloud documentation, so Claude or other third-party models on the same platform stay outside it.

So the $300 is still a Gemini credit, if you call Gemini through the enterprise door. That costs some setup: a Google Cloud project, the Agent Platform API enabled, and service-account authentication instead of an AI Studio key. Google's trial terms also block GPUs, Marketplace purchases and quota increases while you are on the trial, which rarely matters for model calls but does matter if you planned to fine-tune or self-host. For an early team that wants to run a real evaluation on Google's current models with paid-tier data handling, $300 on Agent Platform buys a meaningful amount of testing at the prices in section 10.

How to apply this section: if you have a new Google Cloud account, do not point an AI Studio key at the $300 and expect it to pay. Either stay on the free tier for experiments, or call Gemini through Agent Platform if you want the trial to fund paid-tier usage. And if you are about to apply for the startup program, remember that its Start tier asks that you "Have not received Google Cloud credits beyond free trial", so the trial itself does not count against you.

5. Route 3: the Google for Startups Cloud Program

This is the route that matters. The Google for Startups Cloud Program is the only place a startup can get a five- or six-figure sum that pays for the Gemini API itself, and Google markets it to Gemini builders directly. Its Gemini page tells founders to apply "for the opportunity to receive Cloud credits (from $2k to potentially $350k), to use towards your Gemini API usage as you scale" - Google for Startups. The program's FAQ says the credits "can be used for Google Cloud services such as BigQuery and Gemini" - Google Cloud FAQ, and the official list of offerings the credits apply to includes the service "Gemini API (AEFD-7695-64FA)" with its token SKUs - Google Cloud.

The program has two tiers, and the second has an AI variant. Which one you land in depends almost entirely on how you are funded, so the first decision a founder makes here is not about Gemini at all. It is about whether the company's money counts as "institutional equity" in Google's definition. Our Google for Startups guide covers the whole program, including its non-credit benefits; this section focuses on what reaches the Gemini API and the rules that can stop it.

The Start tier: $2,000 for pre-funded companies

The Start tier gives "$2,000 in Cloud credits", and "Credits are valid for 12 months" - Google Cloud. It is designed, in Google's words, "for digital-native technology startups with a working MVP, a clear business model, and plans to seek venture funding soon". Two hard rules come with it: the company must "Have been founded within the last 24 months" and must "Have not received Google Cloud credits beyond free trial". Alongside the credit, the tier lists "Free tier AI Studio access", 12 months of free Google Workspace Business Plus, $200 in Google Skills training credits and a line worth noticing for this guide: "Program credits cover proprietary Google models".

The 24-month rule is the one that catches people. Older write-ups, including an earlier version of our own, describe a five-year window for the Start tier; Google's pre-funded page now says 24 months, while the five-year window applies to the Scale tier. A bootstrapped company in its third year does not qualify for Start and, without institutional equity, does not qualify for Scale either. For that company the free tier, the trial on Agent Platform and the gateway routes in section 9 are what remain.

What does $2,000 buy? At the prices in section 10, about 380,000 typical requests on Gemini 3.8 Flash at its 2026 introductory price, or roughly 130,000 on Gemini 3.1 Pro Preview. For a pre-funded product that is months of real usage, which is the point: Google wants you building on its models before the round that would put you into the Scale tier.

The Scale and AI tiers: up to $350,000 over two years

The Scale tier is for startups that "Have received recent startup equity funding from pre-seed to Series A by an institutional investor", were founded within the last five years and "Have not received more than $5,000 in Google Cloud credits" - Google Cloud. Google's footnote defines recent: "Pre-Seed/Seed rounds within the last five years, and Series A rounds within the last 12 months", and "Simple Agreement for Future Equity (SAFE) Agreements are eligible". It also lists what does not count: "Funding through private equity, government innovation grants, prize funding, crowdfunding, angel, and friends and family funding will not qualify a startup for scale."

The money arrives in two different shapes. Google's benefits page reads: "Year 1: 100% up to $100,000 USD in Google Cloud credits", "For AI-first startups: 100% up to $250,000 USD in Google Cloud credits", and "Year 2: 20% up to an additional $100,000 USD in Google Cloud credits" - Google Cloud. Year one is a grant: Google pays your whole bill up to the ceiling. Year two is a 20% rebate: Google pays one dollar in five of what you spend. To collect the full second-year $100,000 you would have to spend $500,000 on Google Cloud that year.

That rebate structure changes how to read the famous $350,000. For an AI-first startup, the realistic value is the first-year $250,000, which pays 100% of usage; the year-two rebate only becomes large for a company already spending hundreds of thousands a year. Most seed-stage AI companies will never spend $250,000 on Google Cloud in twelve months, which means the binding constraint on the grant is not its size but how much of it you can use before it expires. Plan your Gemini spend against year one.

The AI variant has its own test. Google describes it as for "Startups that use or plan to use Gemini to deploy AI services as a foundation of their primary product or solution offering" - Google Cloud, and its FAQ adds that inclusion "requires being an AI-first startup (having AI built into your startup's foundational structure and strategy and not only as a productivity/bolt-on layer)". Note the wording of the first sentence: it names Gemini. A company whose application describes a product built on another provider's model is asking Google to subsidize a competitor, and should not be surprised by the standard Scale tier instead.

The Scale tier carries other benefits that matter less here: "$12,000 in Enhanced Support credits for one year" - Google Cloud, free Workspace Business Plus, a "$600 monthly credit for any Google Maps service for one year" and $500 in Google Skills credits. Its partner perks are a separate catalog, and one of them is a Gemini route in its own right: Scale-tier members can receive "$1k in OpenRouter credits for use on Google models, valid for 6 months, along with 0% processing fees for 12 months" - Google Cloud Startup Perks.

What the credits do not cover

Program credits stop at Google's own products. "Program credits cover Google's state-of-the-art models like Gemini and Gemma. Note: Third-party models are billed directly and are not covered by the program credits" - Google Cloud. The FAQ extends this to "any third-party services or offerings including those on Google Cloud Marketplace" - Google Cloud FAQ. Running Claude or another partner model on Google's platform therefore bills your card, with one narrow exception the perks page describes: Scale and Scale AI members "may receive $10,000 USD of credits that can be used toward partner models in Model Garden (including Anthropic, Mistral, AI21)" by asking their Google Cloud Account Executive - Google Cloud Startup Perks. Our Claude credits guide covers that perk from the Claude side.

The exclusions also run the other way, toward who can apply. Google "may not accept companies that have IPO'd or been acquired, educational institutions, government entities, nonprofits, personal blogs or content, dev shops, consultancies, agencies", along with crypto mining and token-distribution companies. An agency building Gemini apps for clients is a common case that does not qualify, however much Gemini it uses.

The Prepay catch: credits wait for cash

This is the rule that decides whether a grant works on the Gemini Developer API, and it sits in a note on the billing page rather than anywhere in the startup program's materials.

Read the note slowly, because it creates a failure that looks like a bug. Your startup credits are applied first, before your own money, so a $10 prepaid balance can sit almost untouched while thousands of dollars of credits pay for traffic. But the credits only flow while that balance is above zero, and if it reaches zero, "all API keys in all projects linked to that billing account will stop working simultaneously" with an HTTP 402. A support reply on Google's forum to a developer in exactly this state put it simply: "you'll need to make a small prepayment to activate the prepayment account and use the credits" - Google AI Developers Forum.

Google's billing page also notes that "Prepay and Postpay billing plans start taking effect March 23, 2026", that older accounts were evaluated and assigned a plan, and that some newer accounts "may be assigned a Postpay account initially". So not every reader is on Prepay today. The safe setup assumes you are, and costs very little.

Applying, and setting up billing before the credits land

The application asks for one thing most founders do not have ready: "Have your 18-character Google Cloud billing account ID and ensure the business email in your application matches your startup's public website domain" - Google Cloud FAQ. The FAQ adds that "your company website domain must match your email domain and the email domain associated with your billing account", that funding is verified through "links to publicly available, online sources detailing the equity investments made in your company", and that most decisions come "within a few business days", with manual reviews taking "10 days or more".

The billing account ID is where the credits land, which ties this section back to the usage tiers in section 2. Credits deposited on a brand-new billing account sit behind that account's Tier 1 cap of $250 a month on the Gemini API. The practical sequence is short, and doing it in this order avoids the two most common stalls.

1. Pick the billing account your Gemini projects will use, on your company domain
2. Link your AI Studio project to it and buy the $5 minimum of Prepay credit
3. Turn on auto-reload with a small reload amount and a monthly limit
4. Apply with that billing account ID and public links proving your round
5. Check your usage tier in AI Studio once credits land, and plan around its cap

The order matters because each step removes a failure. A domain mismatch gets the application rejected before anyone reads it. A missing Prepay balance leaves approved credits idle. A missing auto-reload turns a $10 balance into an outage the first busy week. And the tier check tells you whether your launch fits inside the cap; if it does not, the alternative is Agent Platform, which bills through ordinary postpaid Cloud Billing and where the program's own pages say credits "cover Google's state-of-the-art models like Gemini". Google's onboarding guide for program members points the same way: "To use the Gemini API with enterprise-level support, use Gemini via Vertex AI in the Google Cloud console" - Google for Startups onboarding guide.

How to apply this section: if you have institutional equity and an AI-first product, apply for the Scale AI tier now and treat year one as the prize. If you are pre-funded and under two years old, take the Start tier's $2,000, but know that the Scale tier requires that you have received no more than $5,000 in Google Cloud credits, which the Start tier stays under. Either way, set up Prepay with auto-reload before approval, so the credits work the day they arrive. Our Google for Startups Cloud Program page tracks the current terms.

6. Route 4: accelerators, the AI Futures Fund and investor packages

Beyond the open application, Google channels credits through people it already works with: its own accelerator cohorts, its AI Futures Fund, and the venture firms and accelerators it partners with. These routes rarely publish a fixed amount, and most are selective, but they share one structural advantage over the open program. They come with a relationship, an account team or a partner who can unstick a billing problem, which matters more than founders expect once credits and Prepay start interacting.

None of these routes is a reason to join an accelerator or take money from a particular fund. They are a reason to ask, once you are in one, what Google package comes with it. Many founders hold an investor relationship that would unlock extra credits and never ask.

Google for Startups Accelerators

Google's accelerators are equity-free cohort programs run region by region. The program page lists "Equity-free support for the duration of the program", mentoring from Google teams, early access to AI products, and under credits: "Startups interested in utilizing Cloud credits during the Accelerator may apply here, subject to eligibility review and approval" and "Eligible startups receive product credits" - Google for Startups. No page we read states an amount.

The timing is the practical problem. Every regional accelerator page we checked on October 12 read "Applications are now closed". The North America program shows the calendar a founder should plan around: applications opened May 4, 2026 and closed June 30, 2026, with a September kick-off and November graduation - Google for Startups. Its FAQ says it is "best suited for AI/ML-focused Pre-Seed to Series A revenue generating tech startups" headquartered in the United States and Canada, with "a minimum of 5+ employees". If you would qualify, register your interest now; the next window is likely months away, and Google's own Gemini API Sprints page currently says "There are no Gemini API Sprints available at the moment."

The AI Futures Fund

The AI Futures Fund is Google's own vehicle for AI startups, launched by Google Labs on May 12, 2025 and "open to companies anywhere Gemini is available" - Google. Its page lists four benefits: "early access to some of our most advanced AI models from Google DeepMind like Gemini, Veo and Imagen", collaboration with Google DeepMind and Google Labs teams, "the opportunity to seek direct investment from Google, when you're ready", and "Google Cloud credits and dedicated technical support from our Cloud experts", footnoted "Subject to Google Cloud eligibility requirements and terms" - AI Futures Fund.

The fund publishes no credit amount, and its applications run through a form on that page. Its banners on October 12 showed regional programs in motion, including "AI Futures Fund and Monashees join forces, funding AI pioneers in Brazil. Accepting applications." and a Japan program with KDDI Open Innovation Fund. Treat it as a relationship route for companies whose product depends on Google's newest models, where early access is worth more than the credits.

Investor and accelerator packages

Google also hands credits to startups through partner firms, on top of the open program. Its startups team wrote in January 2025 that "In 2024, we worked with over 300 accelerators worldwide", extending to their startups "dedicated packages of credits" among other benefits - Google Cloud blog. Venture partnerships work the same way. When Google announced its partnership with Lightspeed in April 2025, it included "upwards of $150,000 in cloud credits for Lightspeed's AI portfolio companies, on top of existing credits available to all qualified startups" - Google Cloud blog.

These announcements are dated, and none of the firms publishes current terms we could verify, so this guide lists no amounts for them in its table. The founders' move is simple: ask your lead investor and your accelerator whether they hold a Google Cloud partnership, and if they do, apply through it rather than through the open form. A partner channel can also be the difference between a manual review that takes "10 days or more" and a referral that is already expected.

The Founders Funds belong in the history section now. Google says they "provided more than $58 million to support founders", and editions such as the Black Founders Fund Europe "provided up to $150K in capital, up to an additional $100K in Google Cloud Credits in their first year" - Google for Startups. The pages describe them in the past tense, and the Women Founders Fund page says "There are no Women Founders Funds available at the moment." Watch them, but do not plan around them.

How to apply this section: list every institution you already belong to (your investors, accelerator, university, bank) and ask each one a single question: "Do you have a Google Cloud startup partnership, and what does it include?" It costs an email each, and it is how most of the undocumented credit in this market actually moves.

7. Route 5: hackathons, sprints and challenges

Hackathons are the most Gemini-specific credits that exist, because some of their prizes are Gemini API credits rather than general cloud money. They are also the least plannable: every Google-run competition we found had ended by October 12, 2026, and Devpost listed no open Google or Gemini hackathon when we checked. This section is therefore a record of what these events offered in the past year, so you can recognize the next one and judge whether it is worth a team's week.

The pattern across the year is consistent. Large Google-sponsored hackathons put most of their value into cash prizes, add Cloud credits to the top places, and occasionally give every participant a small credit. Smaller DeepMind-run events pay in Gemini API credits directly. A startup should enter one when the project it would build anyway fits the theme, not to win credits it could get faster from the startup program.

The table shows where the real value sat. The XPRIZE, backed by Google, dwarfs everything else, but it is a competition for teams with months to spare, not a funding route. The All Things Agentic Hackathon is the more useful model for a startup: every entrant received $150 in Google Cloud credits, so a team that builds something it needed anyway collects a small credit regardless of placing. Google's Gemini API Developer Competition, the 2024 contest famous for its DeLorean prize, ended long ago, and guides that still link to it as an open opportunity are stale.

How to apply this section: follow the Google for Startups and Google AI developer channels for the next event, and enter only when the build serves your roadmap. If you are in the Google for Startups program, ask your startup contact about Gemini API Sprints, the program's hands-on build events, which list no dates right now but are part of the program's Gemini kit.

8. Route 6: research, education and nonprofit programs

Google funds Gemini research generously, and almost none of it reaches a startup. These programs are written for universities, and most forbid commercial use. They belong in this guide for one reason: founders who are also academics, or startups spun out of a lab, often qualify for both sides and should know which door is which before they spend.

The cleanest research route is the one with Gemini in its name. The Gemini Academic Program lets "Qualified academic researchers (such as faculty, staff, and PhD students) in supported countries" apply "to receive Gemini API credits and higher rate limits"; the credits are "granted and removed at Google's discretion", and Google reviews applications "on a monthly basis" - Google AI for Developers. It is the only program we found that gives Gemini API credits outright, with no Prepay detour, and it publishes no amount.

Google Cloud's general research program is larger and plainer. It invites researchers to "Apply for $5,000 in research credits" - Google Cloud, for "faculty, PhD students and postdoctoral researchers from accredited higher education institutions and eligible nonprofit research institutions in approved countries", with awards that "expire 365 days from the coupon redemption date" and "may not be used for commercial purposes" - Google for Education. That last clause is the line a spinout must respect: research credits fund the paper, not the product.

Students and nonprofits get less than the name suggests. Google's student page offers training rather than cloud money, "200 free Google Skills credits" that "expire one year after they are issued" - Google Cloud. Google's nonprofit help page points to the standard free tier and free trial rather than a dedicated Gemini credit - Google for Nonprofits Help, and the startup program excludes nonprofits outright.

How to apply this section: if a founder holds an academic post, apply for the Gemini Academic Program for genuine research work, keep its usage in a separate project from the company's, and never route product traffic through it. Everyone else can skip to the gateways.

9. Route 7: gateways, platforms and banking perks

The last group of routes pays for Gemini through a company that resells it. Model gateways sit between your code and the provider, and a credit on their bill pays for whichever model you route to, Gemini included. For Gemini specifically, gateways have a second, less obvious attraction: their credits live outside Google's billing system, so the Prepay floor, the usage tiers and the $250 Tier 1 cap from section 2 do not apply to them. You trade those rules for the gateway's own.

The trade is cheaper than it sounds, because the leading gateways do not mark up Gemini. On the gateways we checked, every Gemini request is served by Google itself: OpenRouter's endpoints for Gemini 3.8 Flash are Google AI Studio and Google's enterprise platform, priced at $0.75 and $3.75 per million tokens - OpenRouter, the same introductory price Google lists. What you pay is the gateway's fee on buying credit, and what you give up is a direct relationship with Google's support and enterprise terms.

OpenRouter

OpenRouter's startup program is the best-known gateway route. "Eligible startups get up to $3,000 in universal inference credits usable across leading frontier and open models with 0% processing fees" - OpenRouter. A second program page states "up to $5,000" - OpenRouter, and the program terms allow "credits in an amount up to $5,000 USD and/or waived platform fees for a period of up to twelve (12) months", noting that credits "may not be combined ("stacked") with other promotions unless explicitly allowed in writing" - OpenRouter terms. Expect $3,000 and treat anything above it as a bonus.

The eligibility list is short and strict: a pre-Series B startup "Building an AI-native product full-time", "Referred by an approved OpenRouter partner", with a public website, a company-domain email, "less than $500 in lifetime OpenRouter spend", and no earlier startup credits. On expiry, "most credits expire 6 months after issuance" while platform fees stay waived for 12 months. On price, OpenRouter's FAQ says "We pass through the pricing of the underlying model providers without any markup, so you pay the same rate as you would directly with the provider", and that it "charges a fee when you purchase credits" - OpenRouter FAQ.

Two partners matter most for Gemini builders. Google itself is one: as section 5 noted, Scale-tier members of Google's program can receive "$1k in OpenRouter credits for use on Google models". Mercury is another: its perk offers "$1K in universal inference credits, valid for 6 months and usable across all 300+ models on OpenRouter", and warns that "Each applicant is allowed to receive startup credits once, even if they are affiliated with multiple partners" - Mercury. That last rule means you pick one partner route, so pick the one with the largest offer you qualify for. Our OpenRouter for Startups page tracks the program. One thing OpenRouter does not offer is a free Gemini model: its free models from Google are Gemma, the open-weights family.

Vercel

Vercel offers two routes. The small one is its AI Gateway free credit: "Free users (those who haven't made a payment) get $5 of credits every 30 days" - Vercel. The free tier "includes a subset of models, not the full catalog" - Vercel docs, and in Vercel's model catalog the Gemini models marked as free-tier were older ones such as Gemini 2.5 Flash and Flash-Lite, not the 3.x Flash or Pro text models. It is a testing allowance, useful mainly because the gateway "charges no markup and no platform fee on tokens".

The large one is Vercel for Startups: "up to $30,000 in Flexible Commitment Amount, Enterprise-tier access, and startup-focused support" - Vercel. The credit page caps the model share: "You may only use up to 50% of your Flexible Commitment Amount towards Vercel AI Gateway usage", for startups accepted on or after August 6, 2026 - Vercel. So up to $15,000 can buy Gemini at Google's list price, the largest non-Google sum in this guide. Eligibility requires affiliation with an approved Vercel partner, a round no later than Series A, an application "within 12 months from their most recent funding round", and a company website with a matching email address. See our Vercel for Startups page for the current terms.

Cloudflare

Cloudflare can route Gemini but its startup credits cannot pay for it. Its startup program gives "up to $350k in credits", but the FAQ lists "AI Gateway: Temporarily not covered by credits. We are actively working to make it accessible" - Cloudflare. The credits do cover Workers AI up to a cap, but its catalog lists Google's open Gemma models, not Gemini - Cloudflare docs. Our Cloudflare for Startups page covers what the program does pay for, which is a lot, just not Gemini tokens.

Banking perks

Business banks pass along Google credit in three different ways, and only one of them is distinctive. Brex's partner perks list Google Cloud with the line "You can have Google Cloud and Firebase costs covered up to $200,000 USD over 2 years" - Brex, which matches the Google for Startups Scale tier in section 5, reached through Brex as a partner rather than a separate pool of money. Ramp's reward page offers "$350 in free credits and free usage of 20+ products on Google Cloud" - Ramp, without saying whether that credit is one Google allows on the Gemini API. And Mercury's Google Cloud perk currently reads "Offer temporarily unavailable. New offer coming later this year" - Mercury.

No bank is worth choosing for a Google credit; the startup program pays the same whether you apply through a bank or alone. Where a bank helps is as a referral source for the gateway routes above, which require one. Our partner perks pages list every offer Ramp and Mercury carry, and our business bank account comparison weighs those perks against fees and features, which is the right order.

How to apply this section: use gateway credits for the work they suit, which is experimenting across models and providers on one balance, and for keeping a small production path alive while your Google credits wait on approval or Prepay setup. If you hold a Vercel partner referral and run on Vercel already, its startup credit is the largest second source of Gemini money available. Ask your bank and accelerator which gateway programs they are partners of before you apply anywhere.

10. What Gemini credits buy: the burn math

A credit is only as large as the work it pays for, and Gemini's price ladder is steep enough that the same grant can last a quarter or several years depending on one model choice. This section prices the credits in requests rather than dollars, using Google's published paid-tier prices on October 12, 2026 - Gemini API pricing. Every figure below uses the global price that both the Developer API and Agent Platform list.

The prices that matter, per million tokens, are these. Gemini 3.1 Flash-Lite costs $0.25 input and $1.50 output. Gemini 3.5 Flash-Lite costs $0.30 and $2.50. Gemini 3.8 Flash costs $0.75 and $3.75 "through December 31, 2026" and then $1.50 and $7.50 "starting January 1, 2027"; Gemini 3.6 Flash carries the same prices. Gemini 3.1 Pro Preview costs $2.00 and $12.00 for prompts up to 200,000 tokens and $4.00 and $18.00 above that. Output prices include thinking tokens, which is the detail most budgets miss: a model that reasons at length before answering bills every reasoning token at the output rate.

A typical request, priced

To turn prices into something a founder can plan with, take a typical assistant request: 4,000 input tokens (a system prompt, some retrieved context and the conversation so far) and 600 output tokens (a short answer plus light reasoning). On Gemini 3.8 Flash at its 2026 price, that request costs $0.00525. On Gemini 3.1 Pro Preview it costs $0.0152, almost three times as much. On Gemini 3.1 Flash-Lite it costs $0.0019.

The spread is eightfold. The same $1,000 buys about 526,000 of these requests on the cheapest Flash-Lite model and about 66,000 on the Pro model. Two bars deserve attention. "3.8 Flash Batch" shows the effect of Google's 50% Batch (and Flex) discount, which brings the strongest Flash model to Flash-Lite economics for any work that can wait. "3.8 Flash 2027" shows the opposite: the introductory price ends on December 31, 2026, and the same credit buys half as much on January 1. A grant approved this autumn and planned at today's Flash price will run out twice as fast next year.

A worked example: a customer support assistant

Consider a startup whose support assistant handles 50,000 conversations a month, with three model calls of the typical size per conversation, or 150,000 requests. Its monthly Gemini bill depends almost entirely on the model, and on which side of the 2027 price change it falls.

Now set those bills against the grants and the caps. A Start tier grant of $2,000 covers about two and a half months of this workload on Gemini 3.8 Flash at today's price, or seven months on 3.1 Flash-Lite. A Scale AI tier's first-year $250,000 covers it for years on any model, so for a company at this volume the grant's size is not the constraint. The usage tier is. At $787.50 a month, the 3.8 Flash bill blows through Tier 1's $250 cap in about ten days, fits inside Tier 2's $2,000 cap, and at 2027 prices still fits. On Gemini 3.1 Pro Preview, $2,280 a month exceeds even Tier 2's cap, so this product would pause every month until its billing account reached Tier 3, which requires $1,000 paid and 30 days from the first successful payment.

That is the arithmetic behind a claim this guide keeps making: for a funded startup on the Gemini Developer API, billing setup decides more than grant size. A large grant on a new billing account is a small grant for its first weeks.

Where the hidden costs sit

Three line items routinely break a credit plan, and none of them is the per-token price. The first is thinking: a model that reasons for 3,000 tokens before writing a 300-token answer bills 3,300 output tokens, so the "600 output tokens" in the example above can be several times larger on hard prompts. Measure your real output token counts on the free tier before you trust any projection, including this one.

The second is search grounding. On the paid tier, Gemini 3 and newer models include "5,000 free search requests per month (shared across all Gemini 3 and newer models), then $14 per 1,000 requests", and Google's footnote adds that "A customer-submitted request to Gemini may result in one or more queries to Google Search. You will be charged for each individual search query performed." One grounded answer that triggers three searches costs $0.042 in search fees past the free allowance, eight times the token cost of the typical request on 3.8 Flash. A research or news product should model search volume separately from tokens.

The third is media. The Nano Banana image models and the Gemini Omni Flash video model have no free tier, and their output is priced far above text. Google prices Gemini Nano Banana 2.1 images at "$0.0336 per 1K image" and "$0.113 per 4K image", and Gemini Omni Flash video at "approximately $0.10 per second" of 720p video. One 4K image costs about as much as twenty typical text requests on 3.8 Flash, and a ten-second clip about as much as 190, so an app that adds generation can multiply its bill with one feature. None of these costs is a reason to avoid the feature; each is a reason to put it in the plan before the grant arrives.

How to apply this section: measure tokens per request on your real prompts, multiply by your expected monthly volume at the 2027 Flash price rather than today's, and compare the result with your usage tier's cap before you compare it with your grant. If the cap is lower than the bill, fix the billing account first, as section 5 describes.

11. Making credits last: model choice, Batch, Flex and caching

A grant's real size is its dollar amount multiplied by your efficiency, and on Gemini the efficiency levers are unusually large. Google sells four ways to run the same model at four different prices, discounts repeated context automatically, and publishes a model ladder whose bottom rung costs a fraction of its top. A team that uses all of them can stretch a $2,000 Start tier grant several times further than a team that sends every request to the strongest model at standard priority.

None of these levers needs a new architecture. They are request parameters, prompt layout and a routing decision, and each one is documented on Google's own pages. The order below runs from the biggest saving to the smallest.

Choose the model per task, not per product

The largest lever is which model answers. Google's own advice for new projects is to "use our latest models: 3.5 Flash-Lite or 3.8 Flash" - Gemini models, and the burn math in section 10 shows why: the Pro model costs roughly eight times as much per typical request as the cheapest Flash-Lite. Most products have a few hard tasks (planning, complex code, long-document reasoning) and many easy ones (classification, extraction, routing, summarizing). Send the easy ones to Flash-Lite and the hard ones to Flash, and reserve Pro for the minority of requests where a test set shows it changes the outcome.

The practical method is an evaluation set. Take a hundred real requests from the free tier, run them on two models, and compare the answers. If Flash matches Pro on 90 of them, routing those 90 to Flash pays for the evaluation within a day of production traffic.

Batch and Flex: half price for work that can wait

Google's pricing card lists "Batch API (50% cost reduction)" for the paid tier - Gemini API pricing, and the Batch rows on every model table confirm it: Gemini 3.8 Flash drops from $0.75 to $0.375 per million input tokens. Batch jobs can take up to 24 hours, which suits evaluations, backfills, nightly enrichment and document processing.

Flex inference is the newer and more useful option for agents. "The Gemini Flex API is an inference tier that offers a 50% cost reduction compared to standard rates, in exchange for variable latency and best-effort availability" - Google AI for Developers. It is synchronous, targets "Minutes (1–15 min target)" of latency, and is switched on by setting the request's service tier to flex. Google pitches it for "Background agents" and "Offline evaluations", and notes that Flex traffic "counts towards your general rate limits" and may be preempted when standard traffic spikes. It is in Preview, so build a fallback to the standard tier for requests that must complete.

The opposite lever costs money. Priority inference is priced at "75-100% more than Standard" on the same comparison table, and Gemini 3.8 Flash's Priority rows show $1.35 and $6.75 per million tokens. Priority is worth paying for user-facing paths with strict latency needs; it is a poor use of a grant anywhere else.

Caching: stop paying for the same prompt twice

Gemini discounts repeated input automatically. "Implicit caching is enabled by default for all Gemini 2.5 and newer models", and "We automatically pass on cost savings if your request hits caches" - Gemini context caching. A cache hit on Gemini 3.8 Flash is billed at $0.075 per million tokens instead of $0.75, a 90% discount. The minimum cacheable prompt is 4,096 tokens for Gemini 3.8 Flash and 3.1 Pro Preview.

To earn the discount, put large, stable content first. Google's guidance is to place "large and common contents at the beginning of your prompt" and to "send requests with similar prefix in a short amount of time". A support assistant that puts its system prompt and policy documents first, and the customer's question last, can serve most of its input tokens from cache. Explicit caches, which you create and manage yourself, add a storage charge per million tokens per hour, so they pay off only for content reused heavily within that time.

Small rules that save real money

A few billing details reward attention. Google does not charge for failed requests: "If your request fails with a 400 or 500 error, you won't be charged for the tokens used. However, the request will still count against your quota" - Gemini API billing docs. On Agent Platform, regional endpoints cost 10% more than the global one, so use global unless data residency requires otherwise. And remember the 2027 price change: if a design only works at today's Flash price, it does not work.

How to apply this section: route by task, move every job that can wait to Batch or Flex, and restructure prompts so the stable part comes first. Together these can cut a typical bill by more than half, which is the same as doubling the grant.

12. Spend caps: protecting the grant from a runaway bill

Credits create a dangerous kind of comfort: when somebody else is paying, nobody watches the meter. Then an agent loops overnight, a retry storm hits a bug, or a scraper finds a public endpoint, and a month of grant disappears by morning. Google's standard budgets do not prevent this. Its documentation is explicit: "Setting an alerts-only budget doesn't automatically cap Google Cloud or Google Maps Platform usage or spending" - Google Cloud documentation.

What does stop spending is a spend cap budget, a newer Cloud Billing feature that covers exactly the services a Gemini startup uses. "When a spend cap budget is enforced, usage of your specified services is automatically paused for the remainder of the budget period unless you manually lift the spend cap", and the eligible services are the Gemini API, Gemini Enterprise Agent Platform, Cloud Run and Cloud Run functions - Google Cloud documentation. Google's Cloud Tech channel published a five-minute walkthrough in September 2026 that sets one up and trips it live on a Gemini API app:

The detail that matters to credit holders

Spend caps ignore your credits. The documentation says "The cost calculations for spend caps are based on gross, estimated costs and don't include savings and credits." A cap of $500 on a project funded entirely by startup credits will still pause that project after $500 of list-price usage, even though you would have paid nothing. That is the right behavior for protecting a grant, but it means the cap should be set at the gross monthly burn of grant you are willing to allow, not at the cash you expect to pay.

Three further limits shape how to use caps. Each cap covers one project and one service, monthly. "The enforcement of spend caps isn't instant and any cost overages are billed as normal", so a cap limits a runaway rather than preventing every dollar of it. And AI Studio's own project spend caps, set on its Spend page "under Monthly spend cap > Edit spend cap", are marked "Experimental", with Google warning that "You are subject to overages for around a 10 minute latency period" and that "Long-running tasks like batch mode completions and agent sessions may incur overages beyond your project spend cap" - Gemini API billing docs.

The billing account's own tier cap from section 2 is a fourth, involuntary limit. It protects you too, but at a level Google chose, and it pauses every project on the account at once.

How to apply this section: put production and experiments in separate projects, set a Cloud Billing spend cap on the Gemini API service in each at a gross monthly figure you can defend, and set a lower AI Studio project cap on the experimental one. Then let the prepaid balance's auto-reload limit act as the last line. Three caps sound like bureaucracy until the night one of them saves a quarter of the grant.

13. Which route fits your startup

Three questions place almost every startup on the right route, and they come in a fixed order. The first is how you are funded, because it decides which tier of Google's program you can enter: institutional equity opens the Scale tier, a company under two years old without it can take the Start tier, and an older bootstrapped company is outside the program altogether. The second is which door you call Gemini through, because the trial credit and the Prepay rule both depend on it. The third is whether your traffic can wait, because Batch and Flex halve the cost of everything that can.

The diagram below turns those questions into a path. It is deliberately simple: a real company often takes two branches at once, applying to the startup program while running on the free tier, and every branch ends in the same place, because no grant lasts if the billing setup throttles it or the bill runs away.

The diagram hides one judgment call: when to apply. A pre-funded company that expects to raise within months faces a small trade-off. The Start tier's $2,000 arrives now, while the Scale tier requires that you have received no more than $5,000 in Google Cloud credits, so taking Start does not block Scale later. The only thing that can block it is a large earlier credit, for example from an accelerator package, which is worth checking before you accept one.

The second judgment call is which door to build on. For most teams the Developer API is the right default, as Google itself says, because it is simpler and its free tier makes development cheap. The case for Agent Platform is specific: you hold trial credit you want to use on Gemini, your Gemini Developer API account sits on a low tier with a launch coming, or you need enterprise controls. Build your model calls behind one internal client so the door is a configuration setting, not a rewrite.

How to apply this section: follow the diagram once, write down the route it gives you and the date you will act on each step, and put the billing setup on the same list as the application. The founders who lose months to Gemini credits are rarely rejected; they are approved and then stalled.

14. Stacking routes, and the order to apply

Gemini routes stack less than Claude or OpenAI routes do, for a structural reason: most of the money is Google's, and Google runs one startup program per company. You will not hold a Start tier grant and a Scale tier grant at once, and the program's own rules make an earlier large credit a reason to refuse a later one. What stacks is Google's money with someone else's: a gateway balance, a platform credit, a hackathon prize.

That leaves three kinds of stack worth building. The first is Google's grant plus its own perk: Scale-tier members can add the $1,000 OpenRouter credit for Google models. The second is Google's grant plus a platform credit, such as Vercel's AI Gateway share, which runs on a different invoice entirely. The third is free capacity alongside paid: the free tier for development and evaluation, the grant for production. One rule caps the gateway stack: OpenRouter gives startup credits once per applicant "even if they are affiliated with multiple partners", so Google's perk and Mercury's perk are alternatives, not additions.

The order follows from the rules. In month one, build on the free tier, create the billing account you will keep, and measure your token counts. As soon as you have an MVP, apply for the Start tier if you are under two years old and pre-funded, after linking your project to that billing account and buying the $5 Prepay minimum. After an institutional round, apply for the Scale or Scale AI tier with links to public sources for the round, and ask your lead investor whether it holds a Google Cloud partnership that adds credits. Whenever a partner offers one, add a gateway or platform credit for multi-model work.

Here is how that plays out for a typical seed-stage company. It prototypes a document-analysis product on Gemini 3.8 Flash's free tier in month one, using public filings as test data. In month two it takes the Start tier's $2,000, prepays $20 with auto-reload, and launches to its first design partners on the paid tier. In month five it closes a seed round led by an institutional fund, and in month six it applies for the Scale AI tier, describing a product built on Gemini. Its billing account now has five months of payment history, so when the larger grant lands it is already past Tier 1, and the credits start paying the same day.

Build your code so the credits can follow the workload. The same Gemini model has the same name on the Developer API and on Agent Platform, but different endpoints and authentication. Keep model calls behind one client with the door, the model and the service tier (standard, Flex or Batch) in configuration. Moving a workload to wherever the credit sits then becomes a setting, which is the only way a stack of credits from different sources stays manageable.

15. Eight mistakes that waste Gemini credits

Most wasted Gemini credit is not spent on bad ideas; it is never spent at all, because a billing rule stopped it from reaching the API. The other half is spent too fast, on the wrong model or tier. Each mistake below appears either in Google's own documentation or in founders' reports on Google's developer forum, usually both.

Read them as a pre-flight check. Four are about billing setup, two about how requests run, and two about timing, and a fifteen-minute review of all eight before launch prevents nearly all of the losses.

1. Expecting the $300 trial to pay for the Gemini API. Since March 2, 2026 it does not, on new accounts. Use the free tier for development, or call Gemini through Agent Platform if you want the trial to pay.

2. Letting the Prepay balance reach zero. Startup credits stop being consumed at $0 and every key on the billing account returns HTTP 402. Keep a small balance with auto-reload on, even while credits pay nearly everything.

3. Landing a large grant on a brand-new billing account. New accounts start at Tier 1, capped at $250 a month on the Gemini API. Apply with the account that already has payment history, or plan the first weeks around the cap.

4. Applying with mismatched domains. Google requires that your website domain match your email domain and the email domain on the billing account. A personal Gmail billing account is the most common way to fail this.

5. Sending every request to the Pro model. Gemini 3.1 Pro Preview costs about three times as much per typical request as Gemini 3.8 Flash and about eight times as much as 3.1 Flash-Lite. Route by task and prove the need for Pro with an evaluation set.

6. Running background work at standard price. Batch and Flex cost half as much. Anything that can wait minutes or hours, from evaluations to enrichment to nightly agents, should not run on the standard tier while you are on credits.

7. Planning at the introductory Flash price. Gemini 3.8 Flash doubles in price on January 1, 2027. A grant budgeted at today's price runs out twice as fast next year.

8. Treating year two of the Scale tier as a grant. It is a 20% rebate on spend, worth $100,000 only if you spend $500,000. Plan your Gemini usage against year one's 100% coverage, and put the expiry date in your calendar the day the credits arrive.

The pattern across all eight is that Google's credits are generous and its billing system is strict, and the strictness is where the money is lost. Our startup credits playbook applies the same pre-flight check to every credit a startup holds, and our AI credits for startups guide compares Google's program with the other model providers'.

16. Frequently asked questions

Can I use Google Cloud's $300 free trial credit for the Gemini API? Not on accounts opened after March 2, 2026. Google's billing FAQ says "starting March 2026, Gemini API usage costs are specifically excluded from the $300 Google Cloud Free Trial program." Firebase's documentation says the same credit can still be used for the Agent Platform Gemini API, formerly Vertex AI.

Do Google for Startups credits work on the Gemini API in AI Studio? Yes. Google invites founders to use the program's credits "towards your Gemini API usage as you scale", and the Gemini API is on the list of offerings the credits apply to. On the Prepay plan, the credits are only consumed while your prepaid balance is above zero.

How much can a startup get for Gemini? Up to $350,000 through the Google for Startups Cloud Program's AI tier, of which up to $250,000 arrives as 100% coverage in the first year and up to $100,000 as a 20% rebate in the second. Pre-funded startups under two years old can get $2,000 through the Start tier.

Is the Gemini API free? Partly. The free tier gives free input and output tokens on the Flash and Flash-Lite models, within limits shown in AI Studio, but Google uses free-tier content to improve its products, and the Pro model has no free tier. AI Studio itself is free of charge in all available regions.

Why did my Gemini API stop working when I still have credits? Most often the prepaid balance reached $0, which stops the credits and returns HTTP 402, or the billing account hit its tier's monthly cap, which pauses all its projects until the next cycle. A spend cap you set yourself can also pause usage, and it counts usage before credits.

Do Google's startup credits pay for Claude or other third-party models? No. Google says "Third-party models are billed directly and are not covered by the program credits". Scale and Scale AI members can request a separate $10,000 credit for partner models through their Google Cloud Account Executive.

Is there a Gemini API credit for students or researchers? For researchers, yes: the Gemini Academic Program grants Gemini API credits and higher rate limits to qualified academics, and Google Cloud offers $5,000 research credits for non-commercial work. Students get Google Skills training credits rather than cloud credit.

Can I get free Gemini through OpenRouter or Vercel? Not the current Gemini models. OpenRouter's free models from Google are Gemma, and Vercel's free AI Gateway credit of $5 every 30 days covers older Gemini models. Both gateways sell the current Gemini models at Google's list price.

17. How we verified this guide

Every figure in this guide was read on its owner's own page on October 12, 2026. We fetched each page directly, checked every quotation against the page text, and kept a figure only when the company that issues the credit states it; numbers that appear only on third-party deal sites were left out. Model names, prices, tiers and caps come from Google's Gemini API models, pricing, rate-limit and billing pages, read the same day. Community reports are quoted from Google's own developer forum and labeled as such. No application was submitted and no account was created to research this guide.

Some things could not be verified, and the guide says so where they matter: the credit amounts in Google's accelerators and the AI Futures Fund, whether credit-funded spend counts toward Gemini API tier upgrades, whether Ramp's Google Cloud credit applies to the Gemini API, and the exact OpenRouter amount, on which its two program pages disagree. While researching, we found that our own Google for Startups guide described the Start tier's age limit as five years; Google's pre-funded page now says 24 months, and we corrected it.

StartupPerks is not endorsed by Google or any other company named in this guide. No program paid to appear here, and this guide contains no referral links. For the same map of other model providers, see our guides to OpenAI API credits and Claude credits, and our startup perks platforms comparison for where else deals like these are collected.

This guide reflects the programs, terms and prices on their owners' pages as of October 12, 2026. Credit programs and model prices change without notice, so confirm the current terms on each linked page before you apply.