LLM APIs without a card: the ten best options
Every service below gives free LLM access with no bank card. Limits are a snapshot from November 2026; details are one click away in the catalog cards.
1. Google Gemini API — the most generous permanent tier
Gemini Flash: ~10 RPM and ~1,500 requests a day, a key from AI Studio in two minutes. Free prompts may be used to improve Google's products.
2. Volcengine Ark / Doubao — up to 5M tokens a day
ByteDance's platform: the biggest daily cap in the catalog. You need to register in the Chinese region, but no card.
3. Qwen Code — 2,000 requests a day
Alibaba's CLI agent on Qwen3-Coder: 2,000 requests a day with an Alibaba Cloud account. Perfect for coding in the terminal.
4. Gemini CLI — 1,000 requests a day
Google's open agent: 1,000 requests a day with an ordinary Google account, no billing setup at all.
5. Groq — LPU speed
Groq: RPM/RPD caps per model, a key right in the console. The fastest inference for open-source models.
6. Cerebras — 1M tokens a day
Cerebras: ~1 million tokens a day on the free plan and 30 RPM. An excellent latency profile for agents.
7. Mistral — the Experiment plan
Mistral: free models with tight rate limits (~1 request/sec). Production use is paid.
8. Z.AI — free GLM-Flash
Z.AI: GLM-4.7-Flash and GLM-4.5-Flash free and indefinitely, with RPM caps.
9. DeepInfra — 50B DeepStart tokens
DeepInfra: the DeepStart program grants ~50 billion tokens for open-source models — the credit is valid for a year.
10. AgentRouter — a $100–200 promo balance
AgentRouter: a starting balance of up to $200 with access to Claude and GPT. A promo balance, with a rotating model catalog.
The fine print
- free tiers cut rate limits hard — build in retries with backoff;
- promo credits expire: don't activate them until you're ready to use them;
- some providers train their models on free prompts;
- Chinese platforms require a CN phone number at sign-up, but no card.
Snapshot 05.11.2026 · the full list is in the catalog, the “no card” filter