Model Routing
How LazyCodex aligns model profiles with Codex roles.
LazyCodex (LZX) aligns model profiles with Codex configuration and agent roles during installation. The applied values come from the bundled model-catalog.json and the models supported by the Codex build in use.
Current Baseline
In 4.12.1, the bundled model-catalog.json centers the current default profiles on gpt-5.5.
| Profile | Model | Reasoning |
|---|---|---|
| default | gpt-5.5 | high |
| plan mode | gpt-5.5 | xhigh |
| worker | gpt-5.5 | high |
| verifier | gpt-5.5 | high |
The model name you see may differ when Codex or OpenAI model availability changes. This page focuses on where LazyCodex uses model profiles rather than on comparing model performance.
Why Role Profiles Exist
Role profiles separate the shape of the work.
- General work follows the default model setting.
- Planning mode can require stronger reasoning settings.
- Worker and verifier profiles separate execution from judgment, so the same result can be checked from another angle.
This pairs with Agent Roles. Even when several roles move at once, Codex configuration can track which model profile each role uses.
User Notes
- The model name shown after installation can differ from this document. In that case, the latest model names supported by Codex and LazyCodex's
model-catalog.jsontake priority. - Model settings affect both quality and speed. Lowering them casually can weaken planning, review, and QA together.
- If something looks wrong, first check installation state and Codex configuration.