OpenAI API Credits for Startups: Every Route in 2026

Where OpenAI API credits actually come from in 2026, which routes need a venture investor and which do not, and how to make a grant last.

OpenAI does give startups free API credits, but its own program publishes no dollar amount and runs almost entirely through venture capital firms. Its OpenAI for Startups page answers the question "Do you offer API credits?" with "Yes! Please reach out to your VC to learn how to unlock OpenAI API credits." For a founder backed by one of OpenAI's partner firms, that is the best route there is. For everyone else, it is a closed door, and most of the articles that rank for this question paper over it with figures that are no longer on the pages they cite.

The good news is that the door marked "OpenAI" is not the only way in. OpenAI's models are sold through Microsoft's Azure, through Amazon Bedrock and through model gateways such as OpenRouter, and each of those has a startup credit program that anyone can apply to. Microsoft's credits cover OpenAI models sold directly by Azure. AWS's Activate terms let its credits pay for third-party models on Bedrock, and AWS names OpenAI among them. A bootstrapped founder with no investor can therefore run GPT models on credits within a week, just not on credits that say "OpenAI" on the label.

This guide lays out every route we could verify on its owner's own page as of September 30, 2026: OpenAI's partner-VC program, its events and research grants, the fintech rewards from Brex and Ramp, the cloud programs from Microsoft, AWS and Google, OpenRouter's gateway credits, the accelerators that bundle OpenAI credits, and OpenAI's data-sharing offer. It explains how OpenAI's billing treats a credit grant, including a usage-tier rule that can throttle a large grant to $100 a month, prices the credits against OpenAI's current API rates so you know what a grant is worth in requests, and ends with a decision framework and the mistakes that waste credits. Every number links to its source.

Contents

1. The short answer: every route at a glance
2. How OpenAI credits work: grants, expiry and usage tiers
3. Route 1: OpenAI for Startups, through a partner VC
4. Route 2: credits at OpenAI events and hackathons
5. Route 3: the Researcher Access Program (up to $1,000)
6. Route 4: Brex and Ramp partner rewards
7. Route 5: Microsoft for Startups and OpenAI models on Azure
8. Route 6: AWS Activate and OpenAI models on Amazon Bedrock
9. Route 7: Google Cloud and OpenAI's open-weight models
10. Route 8: OpenRouter for Startups (up to $3,000)
11. Route 9: accelerators and cohort programs
12. Route 10: complimentary tokens for shared data
13. What the credits buy: the burn math
14. Which route fits your startup
15. Combining routes, and the order to apply
16. Eight mistakes that waste OpenAI credits
17. Frequently asked questions
18. How we verified this guide

1. The short answer: every route at a glance

There are two kinds of "OpenAI credits", and the difference decides most of what follows. The first kind is a credit grant on OpenAI's own platform, spent through api.openai.com and visible on the credit grants page of your OpenAI organization. The second kind is a credit on another company's platform, Microsoft's, Amazon's or OpenRouter's, that happens to pay for OpenAI's models when you call them through that platform. Both buy GPT inference. Only the first is called "OpenAI credits" by anyone, and it is the harder one to get.

The table below lists every route we verified, what its owner publishes about the amount, who qualifies and where the credits are spent. Where the owner publishes no amount, the table says so rather than repeating a figure from a third-party blog.

Read the table from the right-hand column first. If your product must call api.openai.com directly, for example because you rely on a feature that ships there first, only the top five rows help you. If you only need GPT-class models and can point your code at another endpoint, the cloud and gateway rows open up far larger and far more accessible budgets.

The chart shows the shape of the market in one glance: the routes open to anyone top out at a few thousand dollars, and the six-figure routes all run through an investor. Microsoft's investor offer and AWS's Portfolio tier are gated by a referral code or an Organization ID from a backer, exactly like OpenAI's own program. The difference is that the cloud programs publish their ceilings and their eligibility rules, so a founder can plan around them.

2. How OpenAI credits work: grants, expiry and usage tiers

Before chasing credits, it is worth knowing how OpenAI's billing treats them, because three of its rules decide how much of a grant you can actually use. All three are stated in OpenAI's own documentation.

Rule one: new accounts are prepaid. OpenAI's prepaid billing article says "New API accounts use prepaid billing": you buy credits, and usage draws the balance down. The minimum purchase is $5, the default is $10, and auto-reload is switched on by default during setup. When a prepaid balance hits zero, requests start returning a credit_balance_exhausted billing error. The same article explains the order of spending: "If your account has free credits, those are used before purchased credits." OpenAI's help center does not describe an automatic sign-up credit for new API accounts, so plan on buying the first $5 yourself.

Rule two: credits expire, and OpenAI will not extend them. Purchased credits "expire after 1 year," and "OpenAI cannot extend the expiration date of purchased credits or credit grants." That second clause is the one that bites. A grant given for a hackathon, a research project or a partner-VC program comes with an expiry date, and the date is final. OpenAI's startup FAQ points founders to the credit grants page to "track your credit usage," which is where each grant's expiry is shown. Put that date in your calendar the day the grant lands.

Rule three: your usage tier caps monthly usage, and tiers follow purchases, not grants. OpenAI's rate limits guide says each organization has "an approved monthly usage limit," and that "your organization's usage tier upgrades automatically as its total credit purchases reach each threshold." The tiers in September 2026 are:

The practical consequence is easy to miss. A startup that receives a large credit grant on a brand-new organization and has never bought credits sits in the Free tier, where usage is limited to $100 a month. A $2,000 grant consumed at $100 a month takes twenty months, and it may expire first. A $100 purchase moves the organization to Launch and its $5,000 monthly limit, and a $500 purchase moves it to Grow. The purchased credits are not wasted, since they are spent after the grant runs out, but they are the key that lets the grant be used at the speed a real product needs. OpenAI's partner-VC program also lists rate limit upgrades among its benefits, and its VC partnership page names "usage tier upgrades" among the resources partner firms receive, so a founder on that route should ask the partner for the tier upgrade at the same time as the credits.

Rate limits scale with the tier as well. The same guide lists, for the Astra, Sol and Terra model families, 5,000 requests per minute and 1,000,000 tokens per minute on Build, 10,000 and 4,000,000 on Launch, and 15,000 and 40,000,000 on Grow, with higher token limits for the Luna family. A prototype rarely hits these. A launch day does.

One more rule governs what happens at the end. OpenAI's startup FAQ says that once credits run out, "you'll automatically switch to standard billing. No downtime, no surprises." That is convenient for uptime and dangerous for budgets, which is why the mistakes section recommends a spend limit before the grant lands.

3. Route 1: OpenAI for Startups, through a partner VC

OpenAI's own startup program is real, generous in its benefits and deliberately narrow in its entry. The OpenAI for Startups page organizes its offer into build, grow and scale stages, and the scale stage is the one with money in it: "Unlock credits and dedicated support through eligible VC partners." The FAQ spells out the mechanics.

Who qualifies. "Some VCs, like Thrive Capital, Sequoia, a16z, Kleiner Perkins, and Conviction Partners, are part of our partner network. The portfolio companies of these firms get access to additional resources." The five names are examples, not the full list, and OpenAI's advice for everyone else is simple: "If you're not sure whether your VC is a partner, just ask them. They'll know if they're in our network and how to access OpenAI resources."

What it unlocks. A portfolio company of a partner firm "may be eligible for benefits like free API credits, rate limit upgrades, and time with our solutions engineers or other technical team members," plus invitations to exclusive events and "a direct line to the Startup team." The page does not publish a credit amount, and neither does the VC partnership application. Any figure you read elsewhere for this program is either one partner's arrangement or a guess.

How to apply. OpenAI's instruction for credits is to "reach out to your VC to learn how to unlock OpenAI API credits." The FAQ adds that "you won't need a pitch deck," and that OpenAI asks for your organization ID when you apply. You find it by logging in to the OpenAI platform, opening the organization menu, and going to Settings, then Organization settings, where the General tab shows the ID. Create the organization under the company, not under a personal account, before you ask, because the grant lands on whichever organization you name.

If your investor is not a partner. OpenAI runs a VC Partnerships Application for funds: it offers partner firms "OpenAI API credits, usage tier upgrades, exclusive access to events and resources, and more," and it is "intended for venture capital firms backing startups that are actively building with AI." A founder cannot apply on a fund's behalf, but pointing an interested investor at the form costs nothing, and several investors have obvious reasons to join.

Where it fits. This route is the best one available to a VC-backed startup whose product runs on OpenAI's platform: credits land directly on your organization, the rate-limit upgrade removes the tier problem described above, and the technical contact is worth more than the credits for a team pushing the models hard. The same page advertises zero data retention on paid plans for startups that need it for compliance, which matters to the healthcare and financial products that often qualify for venture funding in the first place. Our OpenAI for Startups program page tracks the terms against OpenAI's page, and the StartupPerks matcher shows it alongside every other AI program your startup qualifies for.

4. Route 2: credits at OpenAI events and hackathons

OpenAI also hands out credits in person. The startup FAQ says the organization ID is requested "when applying for credits or at events where we give credits out, like hackathons." That is the whole of what OpenAI publishes about event credits, and it is worth taking literally: credits appear at some events, in amounts set for each event, and the organization ID is what you bring.

The events themselves are listed on the same page. In September 2026 the OpenAI for Startups calendar listed Builder Lounges in Berlin, Stockholm and Paris co-hosted with startups, and a run of university evenings co-hosted with venture firms: MIT with Conviction and with Index Ventures, Harvard with NEA, Yale with Founders Fund, Penn with First Round, Columbia with Thrive Capital, and Brown with Chemistry. The page also points founders to Build Hours, OpenAI's live technical sessions, and to the Startup Drop newsletter, which sends "updates, events, and resources right to your inbox each month."

How to use this route. Treat events as a way to meet OpenAI's startup team and the partner investors in the room, with credits as a possible bonus rather than the goal. A founder who attends a university evening co-hosted by a partner firm is one conversation away from the partner-VC route in section 3, which is where the larger grants live. Subscribe to the newsletter, keep your organization ID handy, and treat any event credit as a prototype budget with a fixed expiry.

5. Route 3: the Researcher Access Program (up to $1,000)

OpenAI's Researcher Access Program is the one OpenAI credit program with a published amount and an open application. It is not a startup program, and it should not be treated as one, but founders who come from research, and research-driven startups studying how their models behave, can qualify on the merits.

What it gives. "Researchers can apply for up to $1,000 of OpenAI API credits to support their work." The credits "are valid for a period of 12 months and they can be applied towards any of our publicly available models." They "expire after 1 year, and cannot be extended or renewed."

Who it is for. OpenAI says it supports "researchers using our products to study areas related to the responsible deployment of AI and mitigating associated risks, as well as understanding the societal impact of AI systems." The page lists research areas that include alignment, fairness and representation, societal impact, misuse potential, human-AI interaction, economic impacts, generalization and transfer learning and multimodal measurements. It encourages "applications from early stage researchers in countries supported by our API" and is "especially interested in subsidizing work by researchers with limited financial and institutional resources."

How and when to apply. The application runs on SurveyMonkey Apply, a third-party service, and asks about your research question and planned use of OpenAI's products. The timing is the part to plan around: "applications are reviewed once every 3 months (in March, June, September, and December)," and grants "take 4-6 weeks to credit to awardees after they receive their application decision." A founder who applies in October waits for the December review and receives credits in January or February. Researchers remain bound by OpenAI's usage policies and its sharing and publication policy, which the page links before you apply.

Who should skip it. A startup whose goal is shipping a product, not studying one, will not fit the program's research areas, and an application dressed up as research wastes a quarter. For product teams, the cloud routes below are faster and larger.

6. Route 4: Brex and Ramp partner rewards

Two corporate-card companies include OpenAI credits in their partner rewards, and the pair shows why a figure has to be read on the owner's page on the day you rely on it.

Brex publishes an amount. On its partner perks page, in a bundle Brex calls the Day Zero Stack ("instant, discounted access to the most powerful AI and startup tools, designed for next-generation founders"), Brex states: "OpenAI: Brex customers can access $1,000 in OpenAI credits for 1 year." Redemption happens inside the Brex dashboard: account and card admins go to Rewards, then Perks and discounts, select the partner's tile, and follow the redemption instructions in the window that opens. Brex notes that "the availability of these perks may change over time," and that a reward removed from the list is no longer available on your account. Read the instructions in the tile before you rely on the credit, so you know which OpenAI account it lands on. Our Brex program page tracks the rest of Brex's partner offers.

Ramp publishes the reward but no longer states an amount. Ramp's OpenAI reward page says: "Startups can receive OpenAI API credits to explore the possibilities of building on top of the OpenAI platform." Ramp lists it under Developer Tools among its partner rewards, which Ramp advertises as "over $350,000 in partner rewards" for Ramp customers. Many guides, and until today our own catalog, describe this reward as "up to $2,500 in OpenAI API credits." That figure is not on Ramp's page as of September 30, 2026, so we no longer repeat it. If the amount matters to your plan, check the offer inside the Ramp dashboard, where redemption happens, before you count on it.

Should you open an account for the credits? Only if you would open it anyway. Both companies are real financial infrastructure with real eligibility rules, and the OpenAI reward is one small line in a large bundle. The right way to decide is the other way round: pick the card and account that fit the company, using our startup banking comparison and the longer startup bank account guide, and then claim every reward that comes with it. Brex's AWS reward is a good example of reading the terms: its page says "All new Brex customers are eligible for $5,000 in AWS credits depending on Activate eligibility," and our AWS guide explains why such partner AWS offers do not stack with each other.

7. Route 5: Microsoft for Startups and OpenAI models on Azure

For a founder without a partner investor, Microsoft is the most accessible way to run OpenAI's models on credits. Microsoft sells OpenAI models through Azure, in the model catalog it calls Microsoft Foundry, and its startup credits pay for them. The terms are published in unusual detail on Microsoft Learn.

The amounts. Microsoft's program overview says startups can "unlock up to $150K in Startup credits over time as your startup demonstrates verified progress, service adoption, and sustained Azure usage," and that startups "backed by a partner in the Microsoft for Startups Investor Network may unlock higher-value benefits earlier." Its FAQ gives the concrete ladder. A startup new to Azure "can start with the Azure startup credit offer: $1,000 for 90 days, then an additional $4,000 for 180 days after business verification." The investor offer, reached with a referral code from an Investor Network partner, "includes Azure credits (most startups begin with $100,000, with the opportunity to earn more)," plus Azure Standard Support and discounts on GitHub Enterprise, Microsoft 365 Business Premium and other Microsoft products.

What the credits cover, precisely. Microsoft's page on sponsorship coverage for Foundry models opens with the rule: "Microsoft for Startups sponsorship credits apply to Microsoft Foundry models that are sold and billed directly by Azure. Models billed by third-party providers, partner services, or available through the Azure Marketplace are not eligible." OpenAI's models on Azure are sold directly by Azure, so the credits apply. The same page lists providers whose popular models are not covered: Anthropic, Cohere (select models), Hugging Face, Meta (select models) and Mistral AI (select models). To see what is covered, use the "Direct from Azure" collection in the Foundry portal. Microsoft sums it up in one line worth remembering: "If Microsoft bills the AI model usage to your Azure subscription, Microsoft for Startups sponsorship credits apply."

Other AI spend the credits cover. Microsoft's guide to using startup credits adds that Fireworks models on Foundry are "billed and sold through Azure," so credits apply to them too, which gives a startup open-weight models alongside OpenAI's under one budget. Kubernetes usage on a sponsored subscription is covered automatically; GitHub usage is covered only after you connect the Azure subscription as a payment method in GitHub.

Eligibility, line by line. The overview lists the criteria: develop a software product or service owned by the company; be privately held and for-profit; be headquartered in a country where Azure is available; have received no more than $350,000 in lifetime free Azure credits; not have raised Series C or later; not be an educational institution, government organization, consultancy or agency; and not be involved in cryptocurrency mining. Applications are made at startups.microsoft.com and "are typically reviewed within three business days."

The fine print that costs founders credits. The FAQ says credits must be activated "within 90 days of accepting the Microsoft for Startups program agreement," or the offer expires and you must reapply. Once activated, credits are "valid for up to two years. Extensions are not available." A credit card is required. Credits "are permanently tied to the subscription you create during redemption" and cannot move to another Microsoft account, so redeem them with the account and subscription the company will keep.

What changes in your code. Credits on Azure are spent on Azure's endpoint for OpenAI models, not on api.openai.com, so your application authenticates against your Azure resource and calls the deployment you create in Foundry. Before you commit, check in the Foundry catalog that the specific model version you need is offered and sold directly by Azure in your region. Our Microsoft for Startups program page tracks these terms, and the AI credits guide ranks Microsoft against the other AI programs.

8. Route 6: AWS Activate and OpenAI models on Amazon Bedrock

AWS now sells OpenAI's frontier models, and AWS startup credits can pay for them. Two facts on AWS's own pages make this route work.

Fact one: OpenAI's models are on Bedrock. AWS's OpenAI on Amazon Bedrock page describes "OpenAI frontier models for agentic coding, autonomous workflows, and complex reasoning, now generally available on Amazon Bedrock," along with Bedrock Managed Agents powered by OpenAI and Codex via Bedrock for development work.

Fact two: Activate credits cover third-party models on Bedrock. Section 1.2 of the AWS Activate Terms says Activate credits "may be applied to offset fees and charges for AWS Marketplace incurred for the use of third-party foundation models available on Amazon Bedrock," which the terms call Bedrock 3P Model Spend. AWS's own explainer on Activate credits names the providers: "AWS Activate Credits can also be used for third-party models available on Amazon Bedrock. These models come from top AI companies, including OpenAI, Mistral AI, Anthropic, and Meta."

The amounts. Our AWS credits guide covers the ladder in depth. In short, Activate Founders starts self-funded startups at $1,000, with select participants eligible for up to $5,000, and Activate Portfolio reaches up to $200,000 for pre-Series B startups with an Organization ID from an Activate Provider (an accelerator, angel or venture firm AWS has approved). Credits "apply monthly to new AWS usage charges on eligible services" and cannot be applied to past bills, so wait for approval before starting an expensive workload.

Why this matters more than it first appears. Portfolio is far easier to reach than OpenAI's partner network: AWS has approved a long list of accelerators and investors as Activate Providers, and many startups already hold an Organization ID through their accelerator's perks portal without having used it. For a startup in that position, a Portfolio grant is the largest pool of credits that can buy OpenAI inference, and AWS credits also cover the rest of the stack the product runs on. Our AWS guide's AI section covers the AI tiers above Portfolio.

What to check before building on it. As with Azure, your code calls Bedrock's API with AWS credentials rather than OpenAI's endpoint, and the models, regions and features on offer are Bedrock's. Confirm on the Bedrock OpenAI page and in your console that the model you need is available in your region, and watch the Credits view in the AWS Billing console to see the grant draw down. Our AWS Activate Founders and AWS Activate Portfolio pages track the current terms.

9. Route 7: Google Cloud and OpenAI's open-weight models

Google Cloud offers OpenAI models too, but only the open-weight ones. Google's documentation for OpenAI models on its platform, now called the Gemini Enterprise Agent Platform (formerly Vertex AI), says "OpenAI models are available for use as managed APIs and self-deployed models." The models listed are gpt-oss-120b and gpt-oss-20b, open-weight language models released under the Apache 2.0 license and described as "well-suited for reasoning and function calling use cases." OpenAI's frontier GPT-6 family is not on Google Cloud.

How this route fits. A startup already on Google Cloud with credits from the Google for Startups Cloud Program can run OpenAI's open-weight models as a managed API priced on Google's own generative AI price list, the kind of first-party usage Google Cloud credits are meant for. Confirm in the Billing console that the model's usage draws on your credit before scaling it, since some offerings in Google's model catalog are billed through the Marketplace instead. Because the weights are open, the same models can also run on any GPU credits you hold, which our AI credits guide covers program by program.

The Google program itself is covered in our complete Google for Startups guide, including the Start and Scale tiers and the AI-first tier, and on our Google for Startups Cloud program page. If your product needs OpenAI's frontier models specifically, Google's credits are the wrong currency, and the Azure and Bedrock routes above are the ones to pursue.

10. Route 8: OpenRouter for Startups (up to $3,000)

OpenRouter is a single API in front of hundreds of models, and its startup credits are "universal": they pay for any model on the platform, OpenAI's included. The OpenRouter for Startups page says "Eligible startups can receive up to $3,000 in universal inference credits, usable across 500+ models available on OpenRouter, along with 0% processing fees for 12 months (including both platform fees and BYOK fees)." OpenRouter's OpenAI provider page lists OpenAI's current GPT-6 and GPT-5.6 models, so the credits buy GPT inference through OpenRouter's endpoint.

Who qualifies. The program "is currently open to pre-Series B venture scalable startups and founders building AI-native products full-time who are referred by an approved OpenRouter partner." Applicants must also have "less than $500 in lifetime OpenRouter spend" and must not have "previously received startup credits." The referral requirement is the gate: like OpenAI's program, it runs through partners, though OpenRouter's partner list and OpenAI's are different, so a startup that misses one may hit the other.

The clock. "If approved, most credits expire 6 months after issuance," while "platform fees will be waived for 12 months." Six months is a short runway for $3,000 of credits at prototype volumes, so this route rewards a team that is already shipping. OpenRouter also supports bringing your own provider keys, at a fee OpenRouter sets at 5% of the normal price of the same model and provider, which the fee waiver covers during its first year.

Why founders use it for OpenAI models. The appeal is optionality. A product built against OpenRouter can switch between OpenAI's models and alternatives from other labs without code changes, which is the strongest hedge a startup has against a price change or a rate limit. Our OpenRouter program page, the most visited program page on StartupPerks, tracks the terms.

11. Route 9: accelerators and cohort programs

Several selective programs bundle OpenAI credits with investment, and one of OpenAI's own programs is a cohort rather than a credit. These are long shots by design, so treat them as upside on top of a route you control.

Neo. The Neo Residency offers startups "$750K uncapped," with Neo taking "participation rights in your next equity round up to 5% ownership," and says "Startups get $450K+ of Azure, AWS, OpenAI, & more," while students in its program "get $100K+ of Azure, AWS, OpenAI, & more." The figure is a bundle across providers, so the OpenAI share is not stated separately.

Conviction Embed. Embed, the program run by Conviction (one of the partner firms OpenAI names), invests "$250K" on an uncapped, no-discount MFN SAFE and lists "$350K in AWS credits, $350K in Azure credits, $500K+ of other compute/inference/hosting credits," naming OpenAI, Anthropic, Baseten, Pinecone, Vercel and Weights & Biases. It describes itself as "highly selective (<1% selection in first batches)" with a class size of 10 to 12 companies.

OpenAI Grove. Grove is OpenAI's program for "technical talent at the very start of their company-building journey," from pre-idea to launch: five weeks of programming, with the first and last weeks in person at OpenAI's San Francisco headquarters (travel covered) and about 15 participants per cohort. OpenAI says "we will offer additional benefits such as credits and introductions to raise capital," without stating an amount. Applications for the latest cohort closed on January 12, 2026, and that cohort ran from January 22 to February 27, 2026. Some guides still quote a fixed Grove credit figure; OpenAI's current page does not, and our accelerators directory lists Grove with the terms as OpenAI now states them.

What is not a credits program. Two other OpenAI-branded programs come up in searches for credits and should be read for what they are. The OpenAI Startup Fund's Converge was an investment program, a six-week cohort whose participants received "a $1 million investment from the OpenAI Startup Fund"; its application page no longer loads as of September 30, 2026. The OpenAI Pioneers Program works with companies in high-impact verticals on domain-specific evaluations and reinforcement fine-tuning with OpenAI's research teams, and does not describe credits. Both are valuable to the right company. Neither is a way to get API credits.

12. Route 10: complimentary tokens for shared data

OpenAI also pays for usage in tokens rather than credits, in exchange for data. Its help article on sharing feedback, evaluation and API data says "Some organizations may qualify for daily complimentary tokens on traffic shared with OpenAI."

How to tell whether you qualify. Open the data-sharing page in your organization's data controls settings. Eligible organizations see "You're eligible for free daily usage on traffic shared with OpenAI"; once enrolled, the page reads "You're enrolled for complimentary daily tokens." OpenAI is explicit: "If you do not see this language on your data sharing page, you are not eligible for the complimentary tokens at this time."

How it works. The tokens apply automatically to traffic you share, only on the projects where sharing is on, and "you need a positive account balance to use OpenAI models and the free tokens." The article describes daily quotas, up to 1 million tokens a day across one group of models and up to 10 million across a group of smaller models, with lower quotas for lower usage tiers, and it warns that a request which would push the day's total past the quota "is billed at normal rates" in full. The article's model list names earlier-generation models, so the offer shown on your own settings page is the authority on which models and quotas apply today.

The trade-off. Sharing means OpenAI receives your inputs and outputs for model improvement. That can be fine for synthetic test traffic or an internal tool, and it is unacceptable for customer data under most contracts and privacy policies. Enable it per project, never on the project that handles customer data, and write the decision into your security documentation so a future enterprise questionnaire does not surprise you.

13. What the credits buy: the burn math

A credit amount means little until it is priced in requests. OpenAI's API pricing page lists prices per million tokens. For short-context requests on the Standard tier in September 2026:

The same page lists Batch and Flex processing at half the Standard price, Fast mode at twice it ("Priority processing was renamed Fast mode on July 30, 2026"), higher prices for long-context requests, and a 10% uplift on regional processing (data residency) endpoints for models released on or after March 5, 2026.

Take a typical application request of 2,000 input tokens and 500 output tokens. On the Standard tier it costs $0.045 on gpt-6-astra, $0.009 on gpt-6.1-sol and $0.00045 on gpt-6-luna. So $1,000 of credits buys about 22,000 requests on Astra, about 111,000 on Sol and about 2.2 million on Luna. The model you choose moves the value of a grant by a factor of 100, more than the difference between the smallest and largest routes in this guide.

The processing tier and caching move it again. The chart below prices a $5,000 grant, the size of Microsoft's self-serve credits or AWS's Founders ceiling, on gpt-6.1-sol under different settings, with the same 2,000-in, 500-out request. "Half cached" assumes half of each request's input is a repeated prefix billed at the cached rate.

The spread is five to one from the same credits and the same model. Fast mode buys speed with half of your requests. Batch and Flex double them for work that can wait, such as classifying a backlog, generating embeddings-adjacent summaries overnight or running evaluations. Prompt caching rewards a stable system prompt and stable context placed at the start of the request. A team that routes background work to Batch, keeps prompts cache-friendly and reserves the frontier model for the requests that need it can make a small grant last longer than a large grant spent carelessly.

Then compare the runway with the expiry. A $1,000 Brex credit valid for one year lasts the year if you spend under about $83 a month. OpenRouter's $3,000, which mostly expires after six months, needs $500 a month to be used in full. A $100,000 Azure investor offer valid for up to two years needs about $4,200 a month. Credits you cannot spend before they expire are worth nothing, so the right target is the grant you can consume, not the largest one on offer.

14. Which route fits your startup

The right route depends on three facts about your company: who backs it, which platform it runs on, and whether it must call OpenAI's endpoint directly. Find the first row that describes you and start there.

Whichever row you start from, add a cloud program for the rest of the stack. The OpenAI-specific routes pay for inference only, while Microsoft's and AWS's credits also cover the databases, storage and compute a product runs on.

Five common profiles, and where each should start.

• The bootstrapped solo founder. Start with Microsoft's self-serve $1,000 and $4,000 on Azure, or AWS Activate Founders if the rest of the product runs on AWS, and prototype on the smallest model that does the job. Buy $100 of OpenAI credits only if a feature requires OpenAI's own endpoint, which also moves that organization to the Launch tier.
• The seed-stage startup with an institutional investor. Ask the investor two questions: is it in OpenAI's partner network, and does it hold a Microsoft Investor Network code or an AWS Activate Provider Organization ID? A yes to any one of them is worth more than every open route combined.
• The accelerator company. Check the accelerator's perks portal for an AWS Organization ID and a Microsoft referral code before applying anywhere, and ask the program whether it can introduce you to OpenAI's startup team.
• The researcher or research-led startup. Apply to the Researcher Access Program with a real research question, timed to the next quarterly review, and use cloud credits for product work in parallel.
• The student builder. Look at programs that name students explicitly, such as Neo's, and at hackathons where OpenAI gives out credits, and keep the organization ID ready for both.

If none of these fits, the StartupPerks matcher ranks every AI and cloud program in our catalog against a description of your startup, and the AI credits ranking compares the model labs' own programs, including Anthropic's Claude for Startups, for teams willing to use more than one model family.

15. Combining routes, and the order to apply

Most of these routes can coexist, because they sit on different billing accounts. An OpenAI credit grant lives on your OpenAI organization, Azure credits on an Azure subscription, Activate credits on an AWS account and OpenRouter credits on an OpenRouter account. Nothing in OpenAI's, Microsoft's, Amazon's or OpenRouter's published terms that we read says holding one disqualifies you from another, and a product that uses OpenAI's endpoint for one feature and Azure's for another can draw on both.

The limits on combining are inside each program. AWS grants Activate credits once per package level, so the many partner offers of "$5,000 in AWS credits" do not add up; our AWS guide explains the one-package rule. Microsoft's eligibility excludes startups that have received more than $350,000 in lifetime free Azure credits. OpenRouter's program is for applicants who have not "previously received startup credits" from it and have spent less than $500 there. Read each program's own rules for itself.

A sensible order of applications:

1. Create the company's accounts first. An OpenAI organization under a company email, an Azure subscription and an AWS account owned by the company, each with a credit card on file and a spend limit set. Grants attach to accounts, and several cannot be moved later.
2. Claim the open, self-serve routes. Microsoft's $1,000 and $4,000 steps, AWS Activate Founders, and any fintech reward from the card you already use.
3. Ask your investors and accelerator the three questions. OpenAI partner, Microsoft Investor Network code, AWS Activate Provider Organization ID.
4. Apply to referral-gated and selective programs where you fit, such as OpenRouter's (with a partner referral) and cohort programs, as upside rather than plan.
5. Revisit when you raise. A funding round is the moment the six-figure tiers open, and the moment to move from self-serve credits to investor-gated ones.

Keep one ledger of every grant, with its amount, its account, its expiry date and the date you plan to exhaust it. With credits scattered across four platforms, the grant that expires unnoticed is the most common loss.

16. Eight mistakes that waste OpenAI credits

1. Planning around a figure the owner no longer publishes. Ramp's reward page no longer states an amount, OpenAI's Grove page no longer states one, and OpenAI's partner program never did. A plan built on a blog's number fails the day the grant arrives smaller. Read the owner's page, and the offer inside the dashboard where you redeem it.

2. Leaving a grant stuck in the Free tier. A new organization that has never bought credits has a $100 monthly usage limit, whatever grant it holds. Buy the $5, $100 or $500 of credits that moves you to the tier your usage needs, or ask the partner program for the rate-limit upgrade it offers.

3. Letting credits expire. OpenAI cannot extend purchased credits or credit grants, Microsoft offers no extensions, OpenRouter's credits mostly expire after six months, and AWS credits expire within one to two years depending on the package. Size the grant to what you can spend before the date, and put the date in the calendar.

4. Assuming every model on a cloud is covered. Microsoft's credits cover models sold directly by Azure and exclude Anthropic's and several other providers' models billed by third parties. Check the "Direct from Azure" collection before building a feature on a model your credits may not pay for.

5. Redeeming on the wrong account. Microsoft's credits are permanently tied to the subscription created at redemption, and OpenAI's grants land on the organization whose ID you submit. Use company-owned accounts from the start.

6. Running up charges before approval. AWS applies credits only to new usage after approval, not to past bills, and every program's grant starts when it is issued, not when you applied. Keep expensive workloads small until the approval email arrives.

7. Sharing customer data for free tokens. The data-sharing offer is for traffic you are willing to give OpenAI. Enable it per project, never on production traffic that carries customer data.

8. Spending frontier-model prices on every request. At September 2026 prices, gpt-6-astra costs about 100 times as much per request as gpt-6-luna. Route simple tasks to small models, send deferrable work through Batch or Flex, and save the largest model for the requests that need it. And set a spend limit before any grant lands, because when credits run out OpenAI switches the account to standard billing automatically.

17. Frequently asked questions

Does OpenAI give free API credits to startups? Yes, through its OpenAI for Startups program, for startups backed by a VC firm in OpenAI's partner network. OpenAI names Thrive Capital, Sequoia, a16z, Kleiner Perkins and Conviction Partners as examples and tells founders to ask their VC how to unlock the credits.

How much are OpenAI's startup credits worth? OpenAI does not publish an amount for its partner-VC program. The published amounts on OpenAI's own platform are the Researcher Access Program's up to $1,000 and Brex's $1,000 for its customers.

Can I get OpenAI credits without venture funding? Not from OpenAI's startup program, which runs through partner VCs. Without funding, the practical routes are Microsoft's self-serve Azure credits ($1,000, then $4,000), AWS Activate Founders ($1,000 to $5,000), rewards from a Brex or Ramp account you already hold, OpenAI's research grants if you do research, and events where OpenAI gives out credits.

Does OpenAI still give free trial credits to new accounts? OpenAI's help center says new API accounts use prepaid billing, with a $5 minimum purchase, and it does not describe an automatic sign-up credit. If free credits are on your account, OpenAI spends them before purchased credits.

Are Azure credits the same as OpenAI API credits? No. Azure credits pay for OpenAI models that Azure sells, called through Azure's endpoint in Microsoft Foundry. OpenAI API credits pay for usage on OpenAI's own platform. Both buy OpenAI models; they are separate accounts and separate bills.

Can AWS credits pay for OpenAI models? Yes, on Amazon Bedrock. AWS's Activate terms let credits cover third-party foundation models on Bedrock, and AWS's explainer names OpenAI among those providers. OpenAI's frontier models are generally available on Bedrock.

Can Google Cloud credits pay for OpenAI models? Only OpenAI's open-weight gpt-oss models, which Google Cloud offers as managed APIs. OpenAI's frontier GPT models are not on Google Cloud.

Do OpenAI credits expire? Yes. OpenAI says purchased credits expire after one year and that it cannot extend the expiration date of purchased credits or credit grants. Research grants expire after one year, and Brex's credit is stated as valid for one year.

Where do I find my OpenAI organization ID? Log in to the OpenAI platform, open Settings, then Organization settings; the General tab shows the organization ID. OpenAI asks for it when you apply for credits and at events where it gives credits out.

What happens when my OpenAI credits run out? OpenAI switches the account to standard billing automatically. On prepaid billing, requests return a billing error when the balance reaches zero unless auto-reload is on, and auto-reload is enabled by default at setup.

Is there a Ramp OpenAI credit of $2,500? Ramp's OpenAI reward page says startups can receive OpenAI API credits but does not state an amount as of September 30, 2026. Check the offer inside the Ramp dashboard before relying on a figure.

Is OpenAI Grove still accepting applications? Not as of September 30, 2026. OpenAI's Grove page says applications for the latest cohort closed on January 12, 2026. It describes credits among the benefits without stating an amount.

Which countries can use these credits? OpenAI's API is available in the countries on its supported countries list, and its research program encourages applicants from those countries. Microsoft requires headquarters in a country where Azure is available. Check each program's own terms for other restrictions.

What is the cheapest way to make credits last? Use the smallest model that does the job, send deferrable work through Batch or Flex at half price, keep prompts cache-friendly, and avoid Fast mode unless latency matters. On September 2026 prices, those choices move the value of the same credits by more than five times.

18. How we verified this guide

Every figure in this guide was read on its owner's own page on September 30, 2026: OpenAI's startup, Grove, Researcher Access, VC partnership, pricing, rate-limit, billing and data-sharing pages; Brex's and Ramp's partner reward pages; Microsoft Learn's program overview, FAQ and Foundry coverage pages; AWS's Bedrock, Activate terms and Activate credits pages; Google Cloud's documentation for OpenAI models; OpenRouter's startup and provider pages; and the program pages of Neo and Conviction Embed. Where an owner publishes no amount, we say so, and where a figure circulating online no longer appears on the owner's page, we do not repeat it.

The same reading corrected three records in the StartupPerks catalog: OpenAI Grove (no stated credit amount, applications closed), Ramp's partner rewards (no stated OpenAI amount) and Microsoft for Startups (the current $1,000 and $4,000 self-serve steps and the $100,000 investor starting point). It also added OpenAI for Startups as a program in its own right. Our catalog of over 1,000 startup programs is re-read against providers' pages on a schedule, and each program page shows when its terms were last checked. For the wider landscape, see how to get startup credits, the best startup cloud credits and the best AI credits for startups.

This guide reflects the terms published by OpenAI, Microsoft, Amazon Web Services, Google, OpenRouter, Brex, Ramp, Neo and Conviction as of September 30, 2026. Programs change amounts, tiers and eligibility often, so confirm the current terms on each official page before you apply. StartupPerks is an independent directory and is not affiliated with OpenAI or any company named here.