Codex CLI — Free LLM API Config
Use OpenAI's Codex CLI with any free LLM backend — no paid OpenAI tier needed.
How It Works
Codex CLI reads OPENAI_BASE_URL and OPENAI_API_KEY from environment variables. Point them at a free OpenAI-compatible backend, and Codex works identically — without the paid API bill.
Quick Config
Groq (fastest, no credit card)
export OPENAI_BASE_URL="https://api.groq.com/openai/v1"
export OPENAI_API_KEY="gsk_your_groq_api_key"
Get key: console.groq.com/keys. Model: llama-3.3-70b-versatile.
NVIDIA NIM (no daily token cap)
export OPENAI_BASE_URL="https://integrate.api.nvidia.com/v1"
export OPENAI_API_KEY="nvapi-your_nvidia_key"
Get key: build.nvidia.com. Model: meta/llama-3.3-70b-instruct.
OpenRouter (one key, 35+ models)
export OPENAI_BASE_URL="https://openrouter.ai/api/v1"
export OPENAI_API_KEY="sk-or-v1-your_openrouter_key"
Get key: openrouter.ai/keys. Model: openai/gpt-oss-120b:free.
Cerebras (gpt-oss with no credit card)
export OPENAI_BASE_URL="https://api.cerebras.ai/v1"
export OPENAI_API_KEY="your_cerebras_key"
Get key: cloud.cerebras.ai. Model: llama3.1-70b.
Persistent Config
# ~/.zshrc or ~/.bashrc
export OPENAI_BASE_URL="https://api.groq.com/openai/v1"
export OPENAI_API_KEY="gsk_your_key_here"
Codex-Specific Model Recommendations
Codex CLI works best with models that support:
| Capability | Recommended Free Models |
|---|---|
| Fast completions | Groq Llama 3.3 70B, Cerebras Llama 3.1 70B |
| Complex reasoning | DeepSeek R1 (OpenRouter, NVIDIA NIM) |
| Long context | Gemini 2.5 Flash (Google AI Studio, 1M ctx) |
| Tool calling | Llama 3.3 70B, Qwen3 Coder 480B |
| Code generation | Qwen3 Coder, DeepSeek R1, GPT-OSS 120B |
Caveats
- Codex CLI expects OpenAI-compatible tool calling. Not all free models support this — Llama 3.3 70B+ and DeepSeek R1 handle it best.
- Set a model name in your Codex config that matches the provider's model ID exactly.
- Rate limits apply to all free tiers.
More Configs
Visit freellm.net/config/codex/ for the interactive config generator.