Free LLM APIs: free tiers, free models and trial credits, measured
· terms checked on 2026-11-07 on each provider’s own page
As of November 07, 2026, of the 41 LLM APIs whose terms we checked, 7 have a free tier or free models with no end date (OpenRouter, Gemini API, Cohere, Novita AI, GroqCloud, Z.AI and SenseNova Token Plan) and 15 give new accounts a one-time credit; 9 have no free access. None of the free options is unlimited: each comes with rate limits, a daily cap or an expiry. Of the free tiers, OpenRouter answered fastest in our probes: 68 ms median time to first byte, averaged over 4 regions.
Which LLM APIs have a free tier that does not run out?
Free tiers and free models with no end date. Each offer links the provider’s own page; next to it is what we measure for that API: median time to first byte averaged over 4 regions, the region it answers fastest from, and the share of host probes answered over the 90 days since 2026-08-09. Rows are ordered by measured speed.
| Provider | What is free | Limits | Average TTFB | Fastest from | Host uptime, 90 d |
|---|---|---|---|---|---|
| OpenRouter free models | model variants whose ID ends in :free (source) | 20 requests a minute; 50 a day, or 1,000 a day once you have bought at least 10 credits | 68 ms | Tokyo 54 ms | 100% |
| Gemini API free tier | free input and output tokens on a limited set of Gemini models (source) | rate-limited; content sent on the free tier is used to improve Google's products; apps for users in the EEA, Switzerland or the UK must use paid services (terms) | 83 ms | US Central 41 ms | 100% |
| Cohere free tier | free evaluation (trial) API keys (source) | 1,000 API calls a month; meant for testing, production needs a paid key | 134 ms | US Central 63 ms | 99.9% |
| Novita AI free models | one model priced at $0 for input and output: Apodex 1.1 Mini (source) | rate limits and how long it stays free are not stated | 179 ms | US Central 93 ms | 99.9% |
| GroqCloud free tier | a Free plan; a payment method is needed only to move up to the Developer tier (source) | per-model rate limits, published on Groq's rate-limits page | 183 ms | US Central 126 ms | 100% |
| Z.AI free models | GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash: input and output both free (source) | rate limits not published in the docs; the other GLM models are paid | 268 ms | Tokyo 94 ms | 99.9% |
| SenseNova Token Plan free tier | a Token Plan at ¥0 a month during the public beta, listed with SenseNova 6.8 Flash Lite and SenseNova U1 Fast (source) | free only while the public beta lasts, which has no announced end, and paid tiers are announced as coming soon; usage capped per rolling 5-hour and weekly window (API docs); up to 20 API keys; mainland China platform | 458 ms | Europe 359 ms | 99.7% |
Which LLM APIs give free credits to new accounts?
A one-time credit or token quota that runs out or expires. Useful to try a provider, not to run on.
| Provider | Credit | Conditions | Average TTFB | Fastest from | Host uptime, 90 d |
|---|---|---|---|---|---|
| Fireworks AI trial credit | a $1 starter credit (source) | eligibility and expiry not stated | 79 ms | Europe 16 ms | 100% |
| Baseten trial credit | credits for new workspaces (source) | amount not stated; Baseten has no free tier or perpetual free plan | 121 ms | US Central 47 ms | 99.9% |
| Cerebras trial credit | a one-time $5 promotional credit (source) | you must add a payment method (not charged); expires 30 days after activation; no free tier that renews | 165 ms | US Central 73 ms | 99.9% |
| Perplexity API trial credit | a one-time $10 signup credit for some new accounts (source) | not every account gets it; it expires 60 days after it is added | 176 ms | US Central 84 ms | 99.9% |
| Anthropic API trial credit | a small amount of free credit for new users (source) | amount and expiry not published | 178 ms | US Central 95 ms | 100% |
| Replicate trial credit | free runs of select models for a while (source) | how much, for how long and which models is not stated; then billing must be set up | 219 ms | US Central 94 ms | 100% |
| Baichuan AI trial credit | ¥80 of free credit for new accounts (source) | valid 3 months from registration; the offer is dated May 2024 on the price page | 230 ms | Tokyo 142 ms | 99.6% |
| SiliconFlow trial credit | $1 in free credits (source) | conditions not stated; the free models on siliconflow.cn belong to the separate China platform | 247 ms | US Central 82 ms | 99.8% |
| Nebius Token Factory trial credit | $25 in Token Factory credits for joining the free Nebius Builder Program (source) | sign-up by form and email check, with sanctions screening; one per verified email; the credits expire 90 days after they are issued | 264 ms | Europe 33 ms | 99.9% |
| AI21 Studio trial credit | $10 credit for new accounts, no credit card needed (source) | expiry given as 7 days on the pricing page and as three months in the docs | 302 ms | US Central 235 ms | 100% |
| Upstage trial credit | $10 in free credit on signup (source) | expiry not stated; pay as you go after that | 404 ms | Tokyo 59 ms | 99.9% |
| Sarvam AI trial credit | ₹100 in free credits on signup (source) | one balance for all Sarvam APIs, chat included; prepaid after that | 407 ms | Tokyo 249 ms | 99.9% |
| Alibaba Cloud Model Studio trial credit | a free quota of typically 1,000,000 tokens per model for new users (source) | valid 90 days; only the Singapore region with International deployment scope | 411 ms | Tokyo 167 ms | 99.9% |
| Volcano Engine Ark (Doubao) trial credit | a free inference quota of 500,000 tokens per model for new users (source) | counted per model and shared across the account; validity not stated; mainland China platform | 502 ms | Europe 350 ms | 99.3% |
| Hyperbolic trial credit | a $1 credit for verifying a phone number (source) | a Free Tier account needs the verified phone to use the API at all; a number already verified on another Hyperbolic account does not count; billed per token after that | 563 ms | US Central 250 ms | 100% |
What is the fastest free LLM API?
Among the free tiers above, by median time to first byte over the last 24 hours: Gemini API is fastest from Europe (90 ms) and US Central (41 ms), OpenRouter from Tokyo (54 ms) and São Paulo (62 ms). That is the network and edge part of every request, measured directly from rented VPSes in four regions, not the model’s generation speed. Where we also send real requests with an API key, the median time to the first token was: Gemini API (gemini-flash-lite-latest): 716 ms from Europe, 1817 ms from US Central, 1895 ms from Tokyo, 1261 ms from São Paulo; GroqCloud (openai/gpt-oss-120b): 354 ms from Europe, 1332 ms from US Central, 1408 ms from Tokyo, 729 ms from São Paulo. More on time to first token.
Is there a free, unlimited LLM API?
No. Every free option above is capped — by requests per minute or per day, by tokens, or by an expiry date — and a capped free tier fails the moment traffic grows. Treat it as a way to test a model and a provider’s latency from your region, then move the steady traffic to a paid plan.
Which LLM APIs have no free access?
On their own pages we found no free tier, free model or signup credit for these:
- OpenAI API — paying organizations that opt in to share API traffic with OpenAI may get daily complimentary tokens (source)
- Mistral AI — its free mode covers Vibe and Studio, not the API (source)
- xAI API (source)
- Together AI — the first purchase must be at least $15 (source)
- SambaNova Cloud (source)
- Featherless AI (source)
- FriendliAI (source)
- Reka (source)
- Moonshot AI (Kimi) (source)
We could not confirm either way from official pages for DeepSeek, MiniMax, StepFun, iFlytek Spark, Baidu Qianfan, DeepInfra, inference.net, Nscale, Writer, Aleph Alpha: what they publish is incomplete or contradicts itself, so they are left out of the tables until it is clear.
Which free LLM APIs work inside OpenClaw, Hermes Agent and n8n?
- OpenClaw: Google Gemini, OpenRouter, Groq, Z.AI (GLM) and NovitaAI
- Hermes Agent: Google / Gemini, OpenRouter, z.ai / GLM and NovitaAI
- n8n: Google Gemini, OpenRouter and Groq
How was this checked?
Every offer was read on the provider’s own pricing, docs or help pages on 2026-11-07; the link next to each one goes to that page. Free terms change without notice, so this page is withdrawn if they go 90 days without a re-check. The latency and uptime are our own measurements, the same as on the fastest AI APIs and uptime; the methodology has the rules. There are no partner links on this page. Seen a change we missed? Report it.