What makes a model good for coding work
A coding benchmark score is not sufficient. The model must also work with your tool. Check these four properties before you select a model.
| Property | Why it is important | Minimum for agent work |
|---|---|---|
| Tool calling | Agents such as Cline, OpenCode and Claude Code read files and run commands through tool calls. A model without tool calling cannot do these steps. | Required |
| Context window | An agent sends a system prompt, tool definitions and file contents in each request. This is often 20K to 60K tokens before your question. | 128K tokens |
| Tokens per minute | A large prompt on a tier with a low tokens-per-minute limit fails with a 429 error on the first request. | More than one full prompt per minute |
| Speed | Autocomplete must answer in less than one second. A slow model is not usable for ghost text. | Fast models for autocomplete |
Today the free catalog has 190 models with tool calling and 56 models with a context window of 1M tokens or more. The tables below filter these models by job.
Free models for coding agents
Use these models with agents that edit files: Claude Code, Codex CLI, Cline, Roo Code, Kilo Code, OpenCode and Aider. Each model in the table supports tool calling and has a context window of at least 128K tokens. The order is by measured intelligence.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| GLM 5.3 | UnoRouter | 1M | 1 req/min 1 req/min per model | Yes | No |
| GLM 5.3 Flash | NVIDIA NIM, UnoRouter | 1M | 40 req/min 40 RPM | Yes | Yes |
| Gemini 3.8 Flash | Google AI Studio | 1M | Recurring provider quota | Yes | Yes |
| Qwen3.8 Flash Next | AMD Radeon Cloud | 262K | Recurring daily account allowance | Yes | No |
| Qwen3.8 Flash | ElectronHub | 1M | $0.25/week shared recurring credit | Yes | Yes |
| Gemini 3.7 Flash | Google AI Studio | 1M | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| GPT-5.6 Luna | ElectronHub | 1M | $0.25/week shared recurring credit | Yes | Yes |
| Dots3 Note Preview | Kilo Gateway, OpenRouter | 512K | 20 req/min 50 req/day 50 req/day account-wide (1,000 after a $10 purchase) | Yes | Yes |
| DeepSeek V4 Flash | AMD Radeon Cloud, BazaarLink, ElectronHub | 1M | 10 req/min 50 req/day Shared rate limit | Yes | No |
| Qwen3.8 27B | ElectronHub, Groq, OVH AI Endpoints | 262K | 30 req/min 250 req/day 70,000 tokens/min ~6M | Yes | Yes |
| Gemini 3.6 Flash | Google AI Studio | 1M | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Qwen 3.8 27B | Cloudflare Workers AI | 262K | Shared 10k neurons/day | Yes | Yes |
| Gemini 3.5 Flash | Google AI Studio | 1M | 10 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| MiniMax M3 | ElectronHub | 1M | $0.25/week shared recurring credit | Yes | No |
| Typhoon 2.5 30B A3B Instruct | Typhoon | 131K | Ongoing free research API (shared rate limits) | Yes | No |
Showing 15 of 109 matching free models. See all models.
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Look at the limits column. A model with a high rank and a limit of 20 requests per day is good for planning steps. It is not good for a long agent loop. A model with 40 requests per minute is better for many small edit steps.
Free code-specialist models
Some model families are trained for code. These include Qwen3 Coder, Codestral, Devstral and GPT-OSS. GLM models are general models with strong code results, and they show in the agent table above. The table shows the free offers for these families today. Some offers do not support tool calling. Use those offers for chat and autocomplete only.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| GPT-OSS 20B | Api.Airforce, Cloudflare Workers AI, Groq +4 more | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| GPT-OSS 120B | CLōD, Cloudflare Workers AI, ElectronHub +3 more | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| OpenAI: GPT OSS 20B | ElectronHub | 128K | $0.25/week shared free credits | No | No |
| Mistral: Devstral 2512 | ElectronHub | 262K | $0.25/week shared free credits | No | No |
| Qwen3 Coder 480B A35B Instruct (turbo) | Inferbase | 8K | 5,000 requests/month; 100 requests and $0.50 usage/day shared | No | No |
| Qwen3 Coder | Septor Labs | 128K | Shared daily allowance | Yes | No |
| GPT-OSS 120B (via Groq) | DreamPrompting | 131K | 100 req/min 5,000 req/day 500,000 tokens/day 500K tokens/day · 5,000 req/day | No | No |
| Qwen3 Coder Plus | ACLIDE | - | €20/month shared internal usage credits · no card | Yes | No |
| Qwen3 Coder 30B A3B Instruct | ElectronHub, ModelScope | 262K | 200 req/day Daily login: 200 credits + 50 for linked Alibaba Cloud; 24h expiry; shared; noncommercial experience | Yes | No |
| Devstral 2 | ACLIDE, xKiro | 262K | 500K tokens/day (1M with Telegram) | Yes | No |
| GPT-OSS 20B (via Groq) | DreamPrompting | 131K | 100 req/min 5,000 req/day 500,000 tokens/day 500K tokens/day · 5,000 req/day | No | No |
| Qwen2.5 Coder 32B | Cloudflare Workers AI | 33K | ~10-20M | Yes | No |
| Codestral | DreamPrompting, LLM7, Mistral +1 more | 256K | 100 req/min 5,000 req/day 500,000 tokens/day 500K tokens/day · 5,000 req/day | Yes | No |
| Codestral 2508 | ACLIDE, Api.Airforce, ElectronHub +1 more | 256K | 1 req/min 1,000 req/day 1,000 req/day · 1 req/min | Yes | No |
| codestral-latest | Api.Airforce | 32K | 1 req/min 1,000 req/day 1,000 req/day · 1 req/min | No | No |
Showing 15 of 17 matching free models. See all models.
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Fast free models for autocomplete and small edits
Ghost-text autocomplete sends a request after almost every pause in typing. It needs low latency, not a high rank. FreeLLMAPI serves these clients on /v1/completions, the legacy OpenAI prompt and suffix format. The table shows fast free models with a context window of at least 32K tokens.
| Model | Free on | Context | Free limits (best offer) | Tools | Vision |
|---|---|---|---|---|---|
| Qwen3.8 27B | BlazeAPI, ElectronHub, Gizmo +3 more | 262K | 30 req/min 250 req/day 70,000 tokens/min ~6M | Yes | Yes |
| GPT-OSS 20B | Api.Airforce, Cloudflare Workers AI, Groq +4 more | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| Nemotron 3.5 Lightning | ElectronHub, Gizmo, Kilo Gateway +2 more | 1M | 20 req/min 50 req/day 50 req/day account-wide (1,000 after a $10 purchase) | Yes | No |
| GPT-OSS 120B | Cloudflare Workers AI, ElectronHub, Groq +1 more | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| GPT-OSS 20B (via Groq) | DreamPrompting | 131K | 100 req/min 5,000 req/day 500,000 tokens/day 500K tokens/day · 5,000 req/day | No | No |
| GPT-OSS Safeguard 20B | Groq | 131K | 30 req/min 1,000 req/day 8,000 tokens/min 200,000 tokens/day ~6M | Yes | No |
| Groq Compound Mini | DreamPrompting | 131K | 100 req/min 5,000 req/day 500,000 tokens/day 500K tokens/day · 5,000 req/day | No | No |
| Gemini 3.5 Flash Lite | ElectronHub, Google AI Studio | 1M | 15 req/min 20 req/day 250,000 tokens/min ~3M | Yes | Yes |
| Step 3.7 Flash | ElectronHub | 256K | $0.25/week shared recurring credit | Yes | Yes |
| Gemini 3.1 Flash-Lite | DreamPrompting, Google AI Studio | 1M | 100 req/min 5,000 req/day 500,000 tokens/day 500K tokens/day · 5,000 req/day | Yes | Yes |
Showing 10 of 215 matching free models. See all models.
Live data from the FreeLLMAPI catalog (version 2026.11.19, checked 19 November 2026). Free limits change often. The catalog is checked every day.
Autocomplete uses many requests. Keep it on a model with high per-minute limits. Do not put it on the same small daily quota as your agent. Read Continue with a free API for an autocomplete config.
How to use one key for all coding tools
FreeLLMAPI pools the free tiers of 45 providers behind one key. It speaks the OpenAI, Anthropic, Gemini and Ollama formats. Each coding tool connects to the format it knows.
Install, then connect a tool
curl -fsSL https://freellmapi.co/install.sh | bash
export FREELLMAPI_API_KEY=<unified-key>
npx freellmapi setup-opencode --url http://localhost:3001 --dry-run
npx freellmapi setup-opencode --url http://localhost:3001Replace setup-opencode with the command for your tool, for example setup-cline, setup-aider or setup-codex. The command reads the live /v1/models catalog, makes a backup of the existing config and merges the new settings.
To steer a request without a change on the dashboard, use a routing model id:
auto: follows your active fallback chain.auto:smart: prefers the models with the highest intelligence. Use it for planning and difficult bugs.auto:fast: prefers measured speed. Use it for autocomplete and small edits.auto:<profile-name>: uses a named fallback chain, for exampleauto:coding. Make the profile on the dashboard first.
FreeLLMAPI also reads the X-FreeLLM-Task-Type: code header. Requests with tools or code markers are treated as code by default. For code, the router moves some weight from speed to intelligence.
Limits and how to stay inside them
Free tiers are small. One agent task can send 20 to 100 requests. A single free key stops a long task fast. FreeLLMAPI counts requests and tokens for each key. When a provider sends a 429 error, the router sends the request to the next model in the chain. The X-Routed-Via header shows which provider answered.
- Add keys from four or more providers. More providers give more requests per day.
- Make a coding profile with only tool-calling, long-context models. Use it with
auto:coding. - Keep sticky sessions on. A conversation stays on one model for 30 minutes, which keeps the agent consistent.
- Turn on context handoff with
FREELLMAPI_CONTEXT_HANDOFF=on_model_switch. The next model then gets a short note that it continues a task. - Start a new session between tasks. A smaller context uses fewer tokens per request.
| Problem | Cause | Fix |
|---|---|---|
| The agent stops after one tool call | The model has weak tool calling | Use a model from the agent table above |
429 on the first request | The prompt is larger than the tier's tokens-per-minute limit | Use a model with a larger limit, or let the router fail over |
| "All models rate-limited after N attempts" | Every key in the chain is at its limit | Add keys from more providers, or wait for the cooldowns to end |
| Edits are slow | A large model on a slow free tier | Use auto:fast for small edits |
Frequently asked questions
What is the best free coding model API?
For agent work, use a model with tool calling and at least 128K context. The first table on this page shows the strongest free models that meet both rules today. For autocomplete, use a fast model from the speed table.
Is there a free Codestral API?
Yes. Mistral serves Codestral on its free tier, and other providers also list it. The code-specialist table shows the current free offers and their limits.
Is there a free Qwen3 Coder API?
Yes. The FreeLLMAPI catalog lists free Qwen3 Coder offers from several providers today. Some offers do not support tool calling, so check the Tools column before you use one in an agent.
Can a free model run a coding agent like Cline or Claude Code?
Yes, if the model supports tool calling and has a large context window. Free tiers have rate limits, so add keys from several providers and let FreeLLMAPI fail over between them.
Which free model is best for code autocomplete?
Use a fast model with high per-minute limits. Autocomplete sends many short requests, so speed is more important than rank. FreeLLMAPI serves autocomplete clients on /v1/completions.
Related guides
- How to use Claude Code for free
- Cline, Roo Code and Kilo Code with a free API
- Use Aider with a free LLM API
- Continue with a free LLM API
- Free LLM API with function calling
- Free LLM API with 1 million token context
Sources
- FreeLLMAPI docs: clients and coding agents (checked 20 November 2026)
- FreeLLMAPI docs: routing strategies and response headers (checked 20 November 2026)
- FreeLLMAPI live model catalog (checked 20 November 2026)