GPT-5.3 Codex — Specs, Pricing & Limits

GPT-5.3 Codex is the model actually named for this site's subject — "optimized for agentic coding tasks in Codex or similar environments," in OpenAI's own words — and it's the one this whole site's most-read comparison exists to put next to the CLI's actual default, because the CLI doesn't default to it.

Its published context window is meaningfully smaller than the CLI default's, and its knowledge cutoff sits several months further back. What it offers in return: a per-token price well under half the default's on both input and output, and the only published reasoning-effort control on this site's entire roster — four distinct levels, stated on this model's own page and nowhere else. No other current model states any effort levels at all, published or otherwise.

It sits in the same rate-limit group as the frontier GPT-5.6 models, so choosing it over the CLI's default costs nothing in throughput — the tradeoff is specifically cost against context headroom and recency, not speed.

The long-context repricing rule that governs several other models isn't stated on this one's own page, and that absence is recorded honestly as not published rather than assumed to carry over from a sibling model. No Batch or Flex pricing is published for it either — Standard and Fast are the only two priced tiers this model offers.

The table below is computed directly from this site's facts module: pricing across every tier this model actually publishes, its four reasoning-effort levels, and its rate-limit table.