Codex CLI

Codex CLI only uses the OpenAI Responses API, so it works with the GPT and Grok models on ORYNX.

Configure

model = "gpt-5.5"
model_provider = "orynx"

[model_providers.orynx]
name = "ORYNX"
base_url = "https://api.orynx.dev/v1"
env_key = "ORYNX_API_KEY"
wire_api = "responses"

export ORYNX_API_KEY="sk-orynx-..."
codex

Notes

• Models to try: gpt-5.5 ($1.50 / $9.00), gpt-5.6-sol ($1.20 / $6.00), gpt-6-astra ($3.00 / $15.00) and grok-4.6 ($0.50 / $1.50).
• With approvals_reviewer = "auto_review", Codex sends approval requests to the codex-auto-review model, billed at its own price ($1.50 / $9.00).
• Codex compacts long sessions by asking the current model for a summary, billed like any other request. Codex only uses OpenAI's server-side compaction when the provider name is exactly OpenAI.
• Claude, Gemini, Kimi, GLM and DeepSeek models do not support the Responses API, so Codex cannot use them.