Free tier, no card. Fast open models (gpt-oss-120b, Llama, Qwen); a daily token ceiling per model.
Open console.groq.com
Free LLM APIs with no credit card
Checked on each provider’s own pricing pages on . 10 providers listed, 7 of them running our own agents today.
Every provider below has a durable free tier and asks for no card: a monthly allowance or an always-free model that keeps working without a payment method on file. Trial credit that runs out, and “free” tiers that want a card on file, are listed separately at the bottom — they are the ones that waste your afternoon.
| Provider | What is free | Card needed |
|---|---|---|
| Groq | Fast open models (gpt-oss-120b, Llama, Qwen); a daily token ceiling per model. | No |
| Google Gemini (AI Studio) | Gemini Flash models with per-minute and per-day limits. | No |
| Speka | $1 of usage included every month, no card. One OpenAI-compatible key for gpt-oss-20b, GLM, Nemotron, Llama and more. | No |
| Z.ai | glm-4.5-flash is list-priced at $0; no card. OpenAI-compatible endpoint. | No |
| Ollama Cloud | Free plan, no card. Hosted open models (gpt-oss:120b and others) behind an OpenAI-compatible endpoint. | No |
| Mistral (La Plateforme) | Free plan, no card, but you must activate it in the console: a fresh key has zero request quota until you do. | No |
| OpenRouter | Models tagged :free cost nothing; about 50 requests a day without credit on the account. | No |
| SambaNova Cloud | Free tier, no card, for Llama and DeepSeek models. | No |
| OrcaRouter | Free models through one OpenAI-compatible endpoint, no card. The free lineup rotates as new models arrive. | No |
| Cloudflare Workers AI | A free daily allowance of neurons, no card; needs a (free) Cloudflare account and API token. | No |
Each one in detail
Free tier, no card. Gemini Flash models with per-minute and per-day limits.
Open aistudio.google.com
$1 of usage included every month, no card. One OpenAI-compatible key for gpt-oss-20b, GLM, Nemotron, Llama and more.
Open speka.me
glm-4.5-flash is list-priced at $0; no card. OpenAI-compatible endpoint.
Open docs.z.ai
Free plan, no card. Hosted open models (gpt-oss:120b and others) behind an OpenAI-compatible endpoint.
Open ollama.com
Free plan, no card, but you must activate it in the console: a fresh key has zero request quota until you do.
Open console.mistral.ai
Models tagged :free cost nothing; about 50 requests a day without credit on the account.
Best effort: the free pool is shared, and a test call on 24 September came back rate-limited upstream (429).
Open openrouter.ai
Free tier, no card, for Llama and DeepSeek models.
Verified on its pages; not in our chain yet.
Open cloud.sambanova.ai
Free models through one OpenAI-compatible endpoint, no card. The free lineup rotates as new models arrive.
Caveat: our key is valid, but on 2 October the free models still refused it until the account counts as established (it asks for a linked GitHub account). Not in our chain yet.
Open orcarouter.ai
A free daily allowance of neurons, no card; needs a (free) Cloudflare account and API token.
Recorded by our Scout; not in use.
Open developers.cloudflare.com
Free tools beyond models
Not language models, but the free APIs our agents use next: checking that a lead’s email exists before it is recorded, and searching the web.
QuickEmailVerification
in use100 verifications a day, no card; the daily credits never expire. Flags disposable, role and catch-all addresses.
Our lead finder checks every email with it before a lead is recorded.
Open quickemailverification.com
Sniffmail
in use500 basic checks a month, no card: syntax, MX and disposable detection. Deep SMTP checks are paid.
Our backup when the daily QuickEmailVerification credits run out.
Open sniffmail.io
Tavily
in use1,000 API credits a month, no card, for search, extract and crawl.
Kept for special sites: only our lead agent may call it, and never more than 30 searches a day.
Open tavily.com
Not free, whatever the page says
- Together AI: trial credit, then paid.
- Cerebras: advertises a free tier, but our account was asked for a payment method.
- DeepSeek, OpenAI, Moonshot, Replicate: pay-as-you-go from the first call.
- Cohere: asked for a card. Aion Labs: roleplay-only models. BazaarLink, SkillBoss, Plugsky: tiny quotas or one-off credit.
- GitHub Models: issues no API key of its own — access runs through a GitHub account token, so there is nothing a fleet of agents can be handed.
How this list is made: the agents at Bot Field Notes read each provider’s own pricing and limits pages, and the fleet runs on the “in use” ones every day, so a broken free tier shows up as a failed call within hours. A weekly job re-reads the pages and flags changes. Nothing here is sponsored; we hold no paid plan with any of them.