All cloud and AI credit offers

Who qualifies: Researchers at academic and nonprofit institutions on high-impact science, especially biology

Apply through the form linked from the support article. Anthropic reviews on the first Monday of each month for scientific merit, potential impact, technical feasibility and biosecurity, then applies up to $20,000 in API credits to your account for six months.

Standard models only, no fine-tuning, no Usage Policy exemption. For-profit researchers are not eligible.

Who qualifies: Any Console organization with asynchronous workloads

Submit up to 100,000 requests or 256 MB per batch. Most finish within an hour and all complete or expire within 24 hours. Haiku 5.5 drops to $0.05 and $0.25 per million, Opus 5.5 to $2 and $10. The discount stacks with prompt caching.

Results available for 29 days. Not available on Bedrock or Google managed endpoints. Fast mode cannot be combined with batches.

Who qualifies: Any request with a stable prefix of at least 512 tokens on the 5.5 models

Mark system prompts, tool definitions and documents as cacheable. A 5-minute cache write costs 1.25x input and pays off after one hit; the 1-hour write costs 2x. Cached tokens also don’t count toward your input rate limit on current models.

Minimum cacheable prompt is 1,024 tokens on Sonnet 5 and 4,096 on Haiku 4.5. Fable 5.1 cache reads are 2.5% of input.

Who qualifies: Pre-Series B startups founded within 10 years with an Org ID from an Activate Provider

Portfolio pays up to $200,000 for startups backed by an accelerator, angel investor or VC firm in the Activate Provider network. Credits cover infrastructure and third-party models on Bedrock. Apply with an AWS Builder ID and expect an answer in 5 to 10 business days.

AWS account must be on the Paid plan. Subject to the AWS Promotional Credit Terms, which exclude Marketplace, Route 53 domains, upfront Savings Plan fees and Enterprise Support. Expiration date is shown on the credit when issued.

Who qualifies: Bootstrapped startups, pre-Series B, founded within 10 years, Paid plan account

Activate Founders is the self-serve tier for startups without an investor Org ID. You start with $1,000, and select participants may qualify for up to $5,000. Create an AWS Builder ID with a professional email, describe the product and funding stage, and link your AWS account.

One application per company unless requesting more than previously received. Free plan accounts are ineligible. Subject to the AWS Promotional Credit Terms.

Who qualifies: Faculty, research staff and graduate students at accredited research institutions

AWS Cloud Credit for Research funds public science-as-a-service tools, cloud migration benchmarks and training workshops. Student awards are capped at $5,000 and faculty awards are sized to the proposal. Applications are reviewed on a rolling basis and take 90 to 120 days.

EC2 usage limited to On-Demand and Spot Instances. Research program credits expire one year after issue under the promotional credit terms. No expedited reviews.

Who qualifies: Every AWS account on the Free or Paid plan, no expiry

Lambda includes one million requests and 400,000 GB-seconds of compute per month in its free tier. It sits alongside DynamoDB at 25 GB of storage and CloudFront at 100 GB of transfer out, all available on both account plans without touching the credit balance.

Does not apply to functions using Provisioned Concurrency. Usage above the allowance draws on credits, then bills at on-demand rates on the Paid plan.

Who qualifies: Early-stage startups with a new personal Microsoft account and a credit card

Startups begin with $200 for 90 days, get $4,800 for 180 days after business verification, and can reach $150,000 through verified progress and sustained usage. Investor Network referral codes bring a two-year period. Apply at startups.microsoft.com by selecting Get started.

Work or school accounts aren’t supported. Accounts that used the free trial, Azure for Students or any sponsorship are ineligible. Credits must be activated within 90 days. Unused credits don’t carry forward.

Who qualifies: Students 18 or older at accredited full-time degree-granting institutions, school email verified

Azure for Students loads $100 of credit for 12 months at sign-up with no card, and includes the 12-month and always-free service amounts. The offer renews annually while you remain a student, and the Education Hub adds developer tool downloads.

One subscription per eligible customer, non-transferable. Credit excludes Marketplace products, support plans, Azure DevOps and Visual Studio subscriptions. The subscription is disabled when the credit is spent.

Who qualifies: Any Azure subscription with a Foundry or Azure OpenAI resource

The Batch API returns completions within 24 hours for a 50% discount on Global Standard pricing. GPT-4.1 batch input is $1 and output $4 per million tokens, against $2 and $8 on demand, so a backfill job costs half what it would on demand.

Asynchronous only, results within 24 hours. Rates vary by model and region; the figures shown are for East US 2.

Who qualifies: Every Azure subscription, including pay-as-you-go, with no expiry

Azure Functions gets 1 million requests a month among 65+ always-free services, next to Cosmos DB at 1,000 RU/s with 25 GB, Container Apps at 180,000 vCPU-seconds and App Service at 10 apps with 1 GB storage. A small API and its database can run on this alone.

Usage above the monthly amount bills at pay-as-you-go rates on an upgraded subscription or draws on credit during the free period.

Who qualifies: Any DigitalOcean user with an account; referred person must be a new customer

Click Get your referral link on the referral program page to open Referrals in your account. When a person who signs up through the link reaches $25 in billings, $25 in credit lands on your account. The referral page doesn’t currently state a credit for the new user.

Reward is account credit with no cash value. Accounts created solely to obtain credits are void under the promotional credit terms.

Who qualifies: All accounts, no credit consumed

App Platform hosts 3 static sites at $0 per month and Functions includes 90,000 GiB-seconds of compute monthly with no invocation charge. Usage above the allowance bills at list price. A docs site and a webhook handler fit comfortably.

Allowances subject to change.

Who qualifies: Teams that sign a 12-month contract through sales

12-month reserved rates are $3.26 for H100, $3.40 for H200 and $1.91 for MI300X, against on-demand rates of $4.41, $4.47 and $2.59. Rates are locked for the term and renewals move to the reserved rate in force at that time.

Requires a contractual 12-month commitment. Pricing effective August 1, 2026 and subject to change.

Who qualifies: Any project on a paid tier with a linked billing account

Batch jobs run asynchronously at half the standard rate: Gemini 3.8 Flash drops to $0.375 per million input and $1.875 per million output tokens through December 31, 2026. Enqueued token limits scale by tier, from 3 million at Tier 1 to 1 billion at Tier 3 for 2.5 Flash.

Not available on the free tier. Jobs complete asynchronously and count toward tier spend caps.

Who qualifies: Active Google AI Pro or Google AI Ultra subscribers who activate Developer Program benefits

Since January 27, 2026 Google Developer Program premium benefits come bundled with Google AI Pro and Ultra subscriptions, including monthly Google Cloud credits usable on the Gemini API, Vertex AI and Cloud Run. Activate through the Developer Program benefits page.

Requires a paid Google AI subscription. Credits are issued monthly and follow Google Cloud credit terms. Subscription prices are set by Google and may change.

Who qualifies: All paid tier projects, standard inference

Gemini 3.8 Flash and 3.6 Flash cost $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, with standard pricing of $1.50 and $7.50 from January 1, 2027. Front-load evaluation and backfill work before the step up.

Standard tier prices. Flex, Priority, Batch and caching have their own rates on the same page.

Who qualifies: Pre-seed to Series A startups with institutional equity funding, founded within five years

Scale tier of the Google for Startups Cloud Program. Year one covers 100% of usage up to $100,000, or $250,000 for AI-first companies. Year two covers 20% up to another $100,000. You also get $12,000 in Enhanced Support credits and 12 months of Workspace Business Plus.

No more than $5,000 in prior Google Cloud credits. Email domain must match the company website. Credits exclude Marketplace and third-party models. Acceptance at Google’s discretion.

Who qualifies: Pre-funded startups with a working MVP, founded within the last 24 months

Start tier of the Google for Startups Cloud Program. Along with the credit you get 12 months of Google Workspace Business Plus and $200 in Google Skills credits. Credits cover Google models on Vertex AI, so a Gemini prototype can run on them.

Must not have received Google Cloud credits beyond the trial. A public website describing the business and product is required.

Who qualifies: Every billing account, including after the trial ends

Monthly allowances on more than 20 products that never expire: one e2-micro VM in three US regions, 30 GB of standard disk, 5 GB of Cloud Storage, 1 TiB of BigQuery queries, 2 million Cloud Run requests and more. Usage inside the limits isn’t charged to credits or your card.

Limits are per billing account per month and do not roll over. Google may change them with 30 days of notice.

Who qualifies: Faculty and researchers at eligible academic institutions

Google Cloud research credits are granted by application through the education site. Institutions in over 70 countries have been funded. Use them for compute, storage, BigQuery and Vertex AI tied to a research project.

Reviewed case by case by Google. Credits are applied to a billing account and follow standard Google Cloud credit expiry.

Who qualifies: Developer plan accounts with a payment method on file

Submit a JSONL file of up to 50,000 requests and choose a completion window between 24 hours and 7 days. Batch pricing is 50% lower than synchronous calls, only completed requests are billed, and batch limits sit outside your per-model rate limits.

Does not stack with the prompt caching discount. Input files up to 200 MB. Results stored for 30 days.

Who qualifies: All plans, automatic on gpt-oss-20b, gpt-oss-120b and gpt-oss-safeguard-20b

Prompt caching is automatic when a repeated prefix clears the model threshold of 128 to 1,024 tokens. Cached input tokens are billed at 50% and don’t count toward rate limits, which stretches the Free plan’s 8,000 tokens per minute as well as the Developer bill.

Place static instructions first in the prompt. Not combined with the batch discount.

Who qualifies: Paid Developer or Enterprise accounts

Set service_tier to flex in the request body to get about 10x the on-demand rate limits at on-demand prices. Requests fail fast with a 498 capacity_exceeded error when capacity is short, so add jittered retries.

Best effort tier. Not available on the Free plan.

Who qualifies: Startups whose VC or accelerator is in OpenAI’s partner network

OpenAI offers free API credits, rate limit upgrades and solutions engineer time to companies referred by a partner investor. The application needs a referral code from the VC, a product description, funding details and your org ID.

No public credit amount. Referral code required. Not a self-serve program.

Who qualifies: Researchers with a defined research question and planned use of OpenAI models

Apply through OpenAI’s researcher access form describing your research and intended use. Approved credits are valid for 12 months on any publicly available model and are tied to the account and institution named in the application.

Credits cannot be transferred to another account or institution. Expired credits are not reinstated; reapply instead.

Who qualifies: Any paid organization with asynchronous workloads

Submit up to 50,000 requests or 200 MB per batch and get results within 24 hours at half the standard token price. gpt-5.4 drops from $2.50 and $15 to $1.25 and $7.50 per million. Batch rate limits are separate from live traffic.

Completion window is 24 hours only. Supported endpoints include responses, chat completions, embeddings, completions, moderations and images.

Who qualifies: Any request on GPT-5.6 or later with a stable prefix of 1,024 or more tokens

Caching is on by default. Keep system instructions, tool definitions and documents at the front of the prompt and identical across calls. Cache writes cost 1.25x input for the 30-minute default or 2.5x for the 12-hour retention option.

Earlier models cache at model-specific rates, 25% of input on gpt-4.1 and 50% on gpt-4o.

Who qualifies: Every account, no card required, models with the :free suffix

Any model ID ending in :free bills nothing per token. New accounts get 20 requests a minute and 50 a day across all free variants combined. Free variants rotate and can be retired, so keep a paid model ID in your fallback list.

A negative balance blocks free requests too. Limits are per account, not per model.

Who qualifies: Accounts with at least $10 in lifetime credit purchases

The daily cap on free model variants rises from 50 to 1,000 requests once your account has purchased at least $10 of credits, all time. The balance doesn’t need to be spent. With the 5.5% fee the purchase costs about $10.55 and the higher limit stays for the life of the account.

The 20 requests a minute cap still applies.

Who qualifies: Any account with credits, batch-capable models only

Bundle chat, responses, messages or embedding requests into a batch and one provider runs them within a 24-hour window at generally half the normal per-token rate, sometimes less. In the beta the median batch finished in 7 minutes. Batches through your own key pay only the BYOK fee.

Discount varies by model and provider. Audio, video and the web search plugin are excluded; web search inside a batch bills at standard rates.

Who qualifies: Standard and Business accounts that add their own provider keys

Add your OpenAI, Anthropic, Google or other provider keys under settings and the provider bills you directly while OpenRouter keeps routing and failover. The first $25,000 of list-price inference each month carries no OpenRouter fee; usage above that pays 5% of list price from your credits.

Enterprise allowances are custom. The startup program waives the BYOK fee for 12 months.

Who qualifies: Every API team, automatic on repeated prompt prefixes

Grok 4.7 bills cached input at $0.50 per million instead of $2.00, and grok-4.3 at $0.20 instead of $1.25. Keep system prompts, tool schemas and examples identical and at the front of each request to hit the cache. Cached tokens still count toward tokens per minute.

Rates double for all tokens when a prompt reaches 200k tokens.

Who qualifies: Every API team, same price on all three grok-4.20 variants

For extraction, summaries, routing and most chat traffic, grok-4.3 costs less than half of Grok 4.7 on output tokens and offers a 1M context window. Save Grok 4.7 for the coding and agent steps where the quality gap shows up in your evals.

Prices apply to prompts under 200k tokens. The Batch API takes a further 20% off.

Who qualifies: Any team under Billing in console.x.ai

Under Billing, open API spend management, set a trigger balance, a top-up amount of at least $5 and a monthly maximum. The console warns at 80% of that maximum. Without auto top-up, requests fail at a zero balance; with no cap, a runaway loop keeps buying credits.

Monthly invoiced billing is separate and requires a request to [email protected].