Model Selection & Profiles

Codex supports multiple models through config.toml profiles, and choosing the right model for each task type is a force multiplier. The default model (codex-1) is optimized for code generation and editing. O3 with high reasoning effort excels at architecture decisions, complex debugging, and code review. GPT-4.1 offers a balance of speed and quality for routine tasks. The profile system lets you switch between these without editing config files — just reference the profile name at session start.

Model selection should be driven by the task, not personal preference. Code generation and file editing: codex-1. Multi-file refactoring with complex dependencies: o3 with high reasoning. Quick fixes and formatting: gpt-4.1-mini for speed. Code review in CI: o3 for thoroughness. The cost difference between models is significant — routing routine tasks to cheaper models while reserving expensive models for complex work can cut API spend by 40-60% without quality loss.

# Task-specific model profiles
[profile.default]
model = "codex-1"
approval = "suggest"

[profile.architect]
model = "o3"
reasoning_effort = "high"
approval = "suggest"
# Use for: design decisions, complex refactoring, debugging

[profile.review]
model = "o3"
reasoning_effort = "high"
approval = "suggest"
# Use for: code review, security audit, PR analysis

[profile.quick]
model = "gpt-4.1-mini"
approval = "auto-edit"
# Use for: formatting, simple fixes, documentation

[profile.ci]
model = "gpt-4.1"
approval = "full-auto"
# Use for: automated CI tasks, PR review bots