Migrating: GPT-5.6 Sol to GPT-5.3 Codex

This page was written when GPT-5.6 Sol was the Codex CLI's default and the obvious question was whether to move to the model named for Codex. Both premises have shifted: the default is now GPT-6.1 Sol, and GPT-5.3 Codex is deprecated — in Codex when you sign in with ChatGPT, and since October 1, 2026 on the API too, which shuts it down on April 1, 2027. The move below still works on an API key until then, but it now means migrating onto a model with an end date. Read it as a record of the trade-offs, and look at GPT-6 Sol, OpenAI's recommended replacement, for anything new.

What changes

Per-token price drops on input, cached input and output — GPT-5.3 Codex undercuts GPT-5.6 Sol on all three, even at Sol's promotional rates. The context window shrinks to a fraction of Sol's, which matters if your sessions accumulate file contents or long histories. The knowledge cutoff moves back several months. Effort levels narrow from none–max to low–xhigh, with no default marked.

What doesn't change

Both sit in the same rate-limit table, so throughput is unaffected, and the output ceiling is the same.

What disappears

No Batch or Flex price, no cache-write price and no long-context rule are published for GPT-5.3 Codex. Sol's long-context behavior is documented; don't assume it carries over.

The actual switch

Inside a session, /model changes the active model; for a persistent default, set model in config.toml — or -m on a single codex exec run. If you sign in with ChatGPT, expect GPT-5.3 Codex to be unavailable or on its way out, and compare against GPT-6.1 Sol, which now undercuts GPT-5.3 Codex on output while offering the larger window.