Free AI API credits compared: OpenAI, Anthropic, Gemini, Groq, xAI
Which AI APIs give free credits or a free tier in 2026, the limits, and the cheapest paid entry at OpenAI, Anthropic, Gemini, xAI, OpenRouter and Groq.
If you’re hoping for a pile of free tokens to test with, only three of the six big AI API providers give a new account anything at all, and none of them put a dollar figure on it. Gemini has a real free tier with free tokens, Groq has a Free Plan with rate limits, and OpenRouter gives a small free allowance plus free model variants. OpenAI has a Free usage tier with a $100 monthly ceiling but no stated trial credit, and Anthropic and xAI publish no free credits at all. Everything here was checked on each provider’s own pricing and docs pages on November 2, 2026, and prices can move without notice.
The comparison table
| Provider | Free credits or free tier | Free limits | Cheapest paid entry | Batch discount | Cache read price |
|---|---|---|---|---|---|
| OpenAI | Free usage tier, no stated credit | $100 per month usage cap, approved geographies | $5 of credit purchases moves you to the Build tier | 50% | 10% of input on the GPT-5 family |
| Anthropic | None stated | Evaluation tier with lower limits for new orgs | Haiku 5.5 at $0.10 input, $0.50 output per 1M tokens | 50% | $0.01 per 1M on Haiku 5.5 |
| Gemini | Free tier with free input and output tokens | Per-model limits shown in AI Studio, data used to improve products | 2.5 Flash-Lite at $0.10 input, $0.40 output | 50% | $0.03 per 1M on 2.5 Flash |
| xAI | None stated | Not applicable | grok-build-0.1 at $1.00 input, $2.00 output | 20% | $0.20 per 1M on grok-4.3 |
| OpenRouter | Small free allowance plus :free models | 20 requests per minute, 50 per day on free models | $5 minimum purchase, 1,000 free requests per day after $10 | Varies by upstream model | Varies by upstream model |
| Groq | Free Plan | Per-model limits on the console limits page | gpt-oss-20b at $0.075 input, $0.30 output | 50% | Not listed on the models page |
All prices are per 1 million tokens in US dollars.
OpenAI gives you a Free tier, not a credit
OpenAI now has four usage tiers: Free, Build, Launch and Grow. The Free tier is available to users in an allowed geography and carries a $100 monthly usage limit. Build requires $5 in total credit purchases and lifts the limit to $500 per month, Launch needs $100 and allows $5,000, and Grow needs $500 and allows $200,000. Upgrades are automatic as your purchases cross each threshold.
OpenAI doesn’t publish any free trial credit for new accounts, so we’re not repeating the old $5 or $18 figures that float around forums. Treat the API as prepaid: the spend limits guide lists a credit_balance_exhausted error for organizations that have used up their prepaid credits. Our OpenAI API credits guide covers the billing model in full, and the OpenAI credits page tracks any promotion we can confirm.
The cheapest way in is $5 of credits. At GPT-5 nano prices of $0.05 input and $0.40 output per 1M tokens, that $5 covers tens of millions of input tokens, which is plenty for a classification or extraction prototype.
Anthropic lists no free credits, but gives 50% off batches
Anthropic lists Haiku 5.5 at $0.10 input and $0.50 output per 1M tokens for prompts up to 100K tokens, Sonnet 5.5 at $2 and $10, Opus 5.5 at $4 and $20, and Fable 5.1 at $10 and $50. There’s no free credit for new API accounts. What there is instead is monthly API credit bundled with the consumer subscriptions, $100 per month on Max, $200 per month on Ultra, and up to $500 per month pooled on Team, which is the closest thing to free Anthropic API credits right now.
New organizations start in what Anthropic calls the Evaluation tier, with limits below the published Start tier while account history builds. The Start tier then caps monthly spend at $500, Build at $1,000 and Scale at $200,000, and requests return a 429 when the cap is hit until the first of the next month. Prompt cache reads cost 5% of the input price on Sonnet 5.5 and Opus 5.5 and 10% on Haiku 5.5, 5-minute cache writes cost 1.25 times input, and the Message Batches API cuts both input and output by 50% with most batches finishing in under an hour.
Gemini is the only true free tier with free tokens
Gemini is the one API in this group where you can make paid-quality calls at zero cost. Google marks free-tier access as “free of charge” on Gemini 2.5 Pro, 2.5 Flash, 2.5 Flash-Lite, 3 Flash Preview, 3.1 Flash-Lite, 3.5 Flash-Lite, 3.6 Flash and 3.8 Flash. Gemini 3.1 Pro Preview is paid-only. The trade is data: content sent on the free tier is used to improve Google’s products, so don’t send customer data through it.
Google doesn’t print free-tier rate limit numbers publicly. Limits are per project, visible in AI Studio, reset at midnight Pacific for daily quotas, and are tighter on preview models. Moving to paid means linking a billing account, which puts you in Tier 1 with a $250 billing cap and a $10 per 10 minutes spend limit. Tier 2 needs $100 paid plus 3 days, Tier 3 needs $1,000 plus 30 days.
Paid pricing starts at $0.10 input and $0.40 output for 2.5 Flash-Lite, with 3.6 Flash and 3.8 Flash at a promotional $0.75 input and $3.75 output through December 31, 2026 before doubling. Batch mode is a flat 50% off. One warning: the $300 Google Cloud trial credit can’t pay for the Gemini API in AI Studio. Our Gemini API credits page has the current model list.
xAI is paid from the first token
xAI publishes per-token prices and nothing about free credits or a free tier. grok-4.7 costs $2.00 input, $0.50 cached input and $6.00 output per 1M tokens for prompts under 200K tokens, doubling above that threshold for the whole request. grok-4.3 is $1.25 input and $2.50 output, and grok-build-0.1 is $1.00 and $2.00.
Discounts are smaller than elsewhere. xAI lists a 20% batch discount on several models including grok-4.7 and grok-4.3, priority processing at 2 times standard rates, and a US regional endpoint at 1.1 times. Third-party sites still advertise a $25 signup credit and a $150 monthly data-sharing credit; xAI doesn’t publish either on its official pages, so we’re leaving them out until it does. Check the xAI credits page for anything that changes.
OpenRouter gives a small allowance and a free model pool
OpenRouter says all new users receive a small free allowance to test the platform, without a dollar figure. More useful are the free model variants, any model ID ending in :free. They’re limited to 20 requests per minute and 50 requests per day until you’ve bought at least 10 credits, after which the daily cap rises to 1,000.
Everything else is prepaid. Purchases run between $5 and $25,000 per transaction, refunds can be requested only within 24 hours, and OpenRouter can expire unused credits 365 days after purchase. Bring-your-own-key traffic gets a $25,000 monthly allowance with no fee, then 5% of what the same model would cost on OpenRouter. Per-model prices mirror the upstream providers, so the batch and caching discounts depend on the model you route to. Our OpenRouter credits page explains the free tier flag.
Groq gives a Free Plan with limits instead of dollars
Groq has a Free Plan and a Developer plan. It publishes the Developer plan’s base limits and tells free users to read their own numbers on the console’s limits page, so there’s no free RPM figure to quote. Upgrading to Developer adds higher limits, batch processing and flex processing.
Groq’s prices are the lowest in this comparison for open-weight models: gpt-oss-120b at $0.15 input and $0.60 output, gpt-oss-20b at $0.075 and $0.30. Llama 3.3 70B and Llama 3.1 8B are listed as “Contact Sales” rather than a price. Batch processing is 50% off synchronous pricing with a window of 24 hours to 7 days. See our Groq credits page for the plan comparison.
Why batch and caching discounts beat credits
A free credit is a one-time event. Batch and cache discounts compound on every request, which is why we’d spend more time on them than on hunting for a signup bonus. OpenAI, Anthropic, Gemini and Groq all take 50% off batch jobs; xAI takes 20%. Cache reads cost 10% of input at OpenAI on the GPT-5 family, 5% to 10% at Anthropic depending on the model, and 10% at Gemini 2.5 Flash, though Gemini adds $1.00 per hour of cache storage. On an OpenAI workload that reuses a long system prompt, a 50% batch discount stacked on a 90% cache discount cuts input costs by 95%.
The OpenAI vs Anthropic comparison runs those numbers on a real workload, and our API free tiers hub keeps a running list of every free tier we can verify.
The short version
- Gemini is the only major API with free tokens on current models, at the cost of your data being used to improve products.
- OpenAI’s Free tier caps usage at $100 per month, and $5 of purchases moves you up; no trial credit appears on official pages.
- Anthropic and xAI list no free credits; Anthropic’s subscription plans include $100 to $500 per month of API credit.
- OpenRouter gives a small allowance and free models at 50 requests per day, rising to 1,000 after $10 in purchases.
- Batch discounts are 50% everywhere except xAI at 20%, and cache reads cost 5% to 10% of input at OpenAI, Anthropic and Gemini.
FAQ
Which AI API is free
The Gemini API free tier offers free input and output tokens on Gemini 2.5 and 3 series Flash models through AI Studio. Groq’s Free Plan and OpenRouter’s :free models also cost nothing within rate limits.
Does openai give free api credits
No, not on any official page. New organizations in approved geographies sit in a Free usage tier with a $100 monthly cap, and buying $5 of credits moves them to the Build tier.
Does anthropic have a free tier
No. Anthropic lists no free tier or free credits for the API. Max, Ultra and Team subscriptions include $100, $200 and up to $500 per month of API credit respectively.
How do openrouter free credits work
All new OpenRouter users get a small free allowance, and models with IDs ending in :free can be called at 20 requests per minute and 50 per day. Buying at least 10 credits raises the daily free-model limit to 1,000 requests.
What is the cheapest AI API to start with
OpenAI’s GPT-5 nano has the lowest input price in this comparison at $0.05 per 1M tokens, and Groq’s gpt-oss-20b has the lowest output price at $0.30. Gemini 2.5 Flash-Lite at $0.10 input and $0.40 output is close behind, and its free tier means you can test before paying.