For a coding agent, the most usable free sources are OpenRouter's :free models (1,000 requests a day after a one-time $10 top-up), Gemini's Flash-Lite models and Cloudflare Workers AI. Groq, SambaNova and Cohere suit short chat and demos. GitHub Models was retired on 30 July 2026, and Anthropic, OpenAI and Hugging Face have no ongoing free API tier. Free tiers can log or train on your prompts.
Coding agents and AI features need a model. The tool is often free; the model is where limits bite. Every figure here comes from the provider's own documentation on 3 November 2026, unless marked as a third-party report, and free tiers change often, so check the limit before you build on it.
The 12 picks
Swipe or use the arrows.
Free tier, no cardGoogle Gemini API (AI Studio)
A free tier that needs no billing, available in India. Free models include Gemini 3.8 Flash and the Flash-Lite models, with limits per Google Cloud project. Google does not publish per-model numbers; third-party reports say 20 requests a day for 3.8 Flash and 500 for Flash-Lite. Free-tier prompts may be used to improve Google products, with human review. Gemini 4 Argon has no free tier, and Google deprecates the Gemini 2.5 models on 1 December 2026 (they route to 3.8 Flash and 3.5 Flash-Lite).
Free models, 50 a dayOpenRouter free models
One key for many models. Models ending in :free allow 20 requests a minute and 50 a day; after a one-time purchase of $10 in credits the limit is 1,000 a day. The free list changes weekly, so read it from the API instead of hard-coding a model.
Free, no cardGroq
Very fast open models such as gpt-oss-120b. The free plan allows 30 requests a minute and 1,000 a day, but only 8,000 tokens a minute and 200,000 a day. One large agent prompt can exceed the per-minute cap, so use it for chat and small scripts.
Free, 10,000 Neurons a dayCloudflare Workers AI
Open models called from inside a Worker. Free: 10,000 Neurons a day, which by our arithmetic from Cloudflare's prices is about 120 requests of 3,000 tokens in and 1,000 out on gpt-oss-20b. Cloudflare says it does not train on your content. Some larger models need a paid plan.
Free trial, about 40 a minuteNVIDIA build.nvidia.com
A free hosted API trial with a developer account and many large open models. The commonly reported limit is 40 requests a minute across all models, according to NVIDIA staff on a forum. It is meant for prototyping, not production, and the model list rotates.
Free, 20 requests a daySambaNova Cloud
A free tier with no card: 20 requests a minute, 20 a day and 200,000 tokens a day per model, on models such as gpt-oss-120b and Llama 3.3 70B. Good for a demo, too small for an agent loop.
$5 trial credits, 30 daysCerebras
Extremely fast inference, but no permanent free tier: new accounts get $5 of credits after adding a verified payment method, and the credits expire after 30 days. Only two models, at 5 requests a minute.
Trial key, 1,000 calls a monthCohere trial key
A free evaluation key: 20 requests a minute but only 1,000 calls a month in total, about 33 a day. Strong embeddings and rerank for search and retrieval projects. Trial keys are for evaluation, not for products.
Free Flash modelsZ.ai GLM Flash models
Z.ai lists GLM-4.7-Flash, GLM-4.5-Flash and the vision model GLM-4.6V-Flash as free. Rate limits are not public and the flagship GLM-5 models are paid. Use a Standard API key, not a Coding Plan key.
One-time, about 1M tokensAlibaba Model Studio (Qwen)
A one-time quota of about 1 million tokens per model for 90 days after you first activate the Singapore region. The setting that stops at the quota is off by default, so turn on Free Quota Only first or you will be billed.
Free for you, user-paysPuter.js
A JavaScript library that lets a front-end app call more than 500 models with no API key. Each user pays from their own free Puter allowance, so you pay nothing. Users must sign in to Puter, and heavy use needs a paid plan.
Free, offlineOllama and LM Studio (local)
Run open models on your own laptop with no key and no limits. Ollama serves an OpenAI-compatible API at localhost:11434. A 16 GB laptop can run a model like Qwen3.5 9B (6.6 GB, Apache-2.0); Devstral Small 2 (24B) needs about 32 GB.
Compare in a table
Show the full comparison table
| Tool | Best for | Free plan |
|---|---|---|
| Google Gemini API (AI Studio) | Long-context chat and small apps | Free tier, no card |
| OpenRouter free models | The most flexible free agent backend | Free models, 50 a day |
| Groq | Fast chat and short prompts | Free, no card |
| Cloudflare Workers AI | A private free daily quota | Free, 10,000 Neurons a day |
| NVIDIA build.nvidia.com | Big open models for prototypes | Free trial, about 40 a minute |
| SambaNova Cloud | Quick demos on big models | Free, 20 requests a day |
| Cerebras | Speed demos, a short trial | $5 trial credits, 30 days |
| Cohere trial key | Embeddings and RAG experiments | Trial key, 1,000 calls a month |
| Z.ai GLM Flash models | Light coding help | Free Flash models |
| Alibaba Model Studio (Qwen) | Trying Qwen models once | One-time, about 1M tokens |
| Puter.js | AI web apps with no backend | Free for you, user-pays |
| Ollama and LM Studio (local) | Private use with no limits | Free, offline |
Get started this week
- Create a Gemini API key in AI Studio and send one request with curl.
- Add an OpenRouter key to Cline or OpenCode and check your daily limit.
- Install Ollama, pull a small model and point a tool at localhost:11434.
How to choose
- For a coding agent, use OpenRouter after the one-time $10 top-up, Gemini Flash-Lite or Cloudflare; use Groq, SambaNova and Cohere for short chat and RAG experiments (our judgement from the limits above).
- Free tiers can log or train on your prompts, and Gemini's free tier does with human review. Never paste secrets, client data or student records into them.
- Keep API keys in environment variables on a server; a key shipped in browser code is public. For an app with no backend, use Puter.js.
- Read the free model list at runtime and re-check limits before a demo: models are retired and limits change often.
Free plans change. Limits above were checked on 4 Nov 2026; confirm on each tool's pricing page before relying on them.
Frequently asked questions
Which free LLM API is best for a coding agent?
OpenRouter's free models after a one-time $10 credit purchase give 1,000 requests a day. Gemini's Flash-Lite models and Cloudflare Workers AI also work. Groq's 8,000 tokens a minute is too small for most agent prompts.
Is there a free OpenAI or Claude API?
Neither has an ongoing free API tier. Anthropic says new users may get a small credit to test the API, and OpenAI documents no free credit. Use the free sources above or a local model instead.
Is GitHub Models still free?
No. GitHub retired GitHub Models on 30 July 2026, including its playground, catalogue and inference API. Older tutorials that use it no longer work.
Will free APIs train on my prompts?
Some do. Google says free-tier Gemini prompts may be used to improve its products, with human review. OpenRouter free endpoints may log prompts, depending on your privacy settings. Check each provider before sending anything private.
Are these free APIs available in India?
Google's Gemini API lists India as an available region, and Groq supports accounts with Indian billing addresses. For several others, such as Cerebras, SambaNova and Cohere, we could not confirm India availability, so test sign-up before planning around them.
Learn this with mentors
Programmes for builders and engineers with admissions open: live classes, mentors and project work.
