Migrating: GPT-5.6 Sol to GPT-5.3 Codex

Switching your CLI usage from the default model to the Codex-branded one is a config change, not a code change — but it's worth understanding what actually moves before you make it, since the two models differ on more than just price.

What changes

Per-token price drops meaningfully on both input and output. Context window shrinks — the Codex-branded model's window is noticeably smaller than the default's, which matters if your sessions routinely accumulate a lot of file content or long conversation history. Knowledge cutoff moves backward by several months. And the Codex-branded model publishes explicit reasoning-effort levels, a control the default model's own page doesn't state.

What doesn't change

Rate limits stay in the same published tier group — this migration doesn't cost you throughput. And output-token ceiling is published as identical between the two, so response length limits aren't a factor in the decision either.

What to recheck before switching

If your workload runs long agentic sessions or works against a large codebase, measure how close your typical session actually gets to the smaller model's context window before committing — a session that fits comfortably today may not once you've made the switch. And because the long-context repricing rule is not published for the Codex-branded model at all, don't assume the same repricing behaviour carries over from the default model if you were previously relying on understanding that rule to budget large requests.

The actual switch

Inside a running session, /model changes the active model for that session. For a persistent default, the CLI's config file is the right place to set it — see the config-file reference guide in the CLI section for the exact key.