# Codex

Find configuration guides for using text models in Codex and check the API format your setup requires.

## Configure the Responses connection

The [SeedRouter Codex guide](https://seedrouter.ai/blog/gpt-6-codex) contains the full setup and its recorded test results. Its configuration uses the Responses API and a named model connection in the user-level Codex configuration.

Add the relevant entries to `~/.codex/config.toml`, preserving unrelated settings:

```toml
model = "gpt-6-sol"
model_provider = "seedrouter"

[model_providers.seedrouter]
name = "SeedRouter"
base_url = "https://api.seedrouter.ai/v1"
env_key = "SEEDROUTER_API_KEY"
wire_api = "responses"
```

Then supply your own key through the environment:

```bash
export SEEDROUTER_API_KEY=sk-your-seedrouter-key
codex
```

The linked guide records Codex 0.157.0 behavior. Review its configuration notes when using a different version; do not copy context or reasoning limits from a different model.

## Verify a coding task

Start in a project where you can inspect the diff and run a known check. Ask Codex to read a file, make one bounded change and report the actual check result. Confirm that the selected model handled the session by comparing the configuration with [usage history](https://seedrouter.ai/usage).

The [Kimi coding guide](https://seedrouter.ai/blog/kimi-k3-claude-code-codex) covers another model option. Use each guide's specific configuration and limitations rather than assuming every text model behaves identically in a coding client.

## Account for the full session

Project instructions, tool definitions, repeated context and reasoning contribute to usage. The [context-window guide](https://seedrouter.ai/blog/context-window-vs-max-output-tokens) explains capacity versus output limits; [current prices](https://seedrouter.ai/pricing) provide the rates.

Codex controls local file access and command execution according to its configuration. SeedRouter supplies the model calls. The existing guide's tests remain historical evidence; this directory page does not report a new live coding benchmark.

## Related resources

- https://seedrouter.ai/models/gpt-6-sol
- https://seedrouter.ai/models/gpt-6-astra
- https://seedrouter.ai/models/kimi-k3
- https://seedrouter.ai/blog/gpt-6-codex
- https://seedrouter.ai/blog/kimi-k3-claude-code-codex