GPT-5.3 Codex — Specs, Pricing & Limits
Deprecated on the API on 2026-10-01; it shuts down on 2027-04-01, and the recommended replacement is gpt-6-sol. Codex with ChatGPT sign-in already calls it deprecated, with no date given there.
Specs
| Status | Priced |
|---|---|
| API deprecation | Deprecated 2026-10-01; shuts down 2027-04-01 — OpenAI's recommended replacement: GPT-6 Sol |
| Context window | 400,000 tokens |
| Max output | 128,000 tokens |
| Knowledge cutoff | Aug 31, 2025 |
| Reasoning effort levels | low, medium, high, xhigh |
| Long-context rule | Not published |
| Cache writes | No additional charge — a written token bills as ordinary input |
| Codex with ChatGPT sign-in | Deprecated — replacement: A recommended model; update scripts and config that still name it. |
Pricing per 1M tokens, every published service tier
| Standard — short context | in $1.75 · cached $0.175 · writes — · out $14.00 |
|---|---|
| Fast — short context | in $3.50 · cached $0.35 · writes — · out $28.00 |
Rate limits by usage tier
| Free | Not supported |
|---|---|
| Build | 5,000 RPM · 1,000,000 TPM · 3,000,000 batch queue |
| Launch | 10,000 RPM · 4,000,000 TPM · 200,000,000 batch queue |
| Grow | 15,000 RPM · 40,000,000 TPM · 15,000,000,000 batch queue |
GPT-5.3 Codex is the model named for this site's subject — "optimized for agentic coding tasks in Codex or similar environments," in OpenAI's words — and two things about it changed this autumn. The CLI's default moved to GPT-6.1 Sol, a model that is now cheaper than GPT-5.3 Codex on output tokens. And OpenAI marked GPT-5.3 Codex as deprecated in Codex when you sign in with ChatGPT, without a date there — then, on October 1, 2026, deprecated it on the API as well, with a shutdown on April 1, 2027 and GPT-6 Sol as the recommended replacement. Until then it is still priced under the API's own "Codex" category, and Codex with your own API key can still select it.
Against the current default, it trades a smaller context window and a knowledge cutoff several months older for a lower input rate — but a higher output rate. For reasoning-heavy agentic work, where output and reasoning tokens dominate the bill, that trade no longer runs in this model's favor — and with an API shutdown date set, anything new built on it is something you'll have to move again.
This site once described its effort levels as the only published reasoning-effort control on the roster. That was wrong even then: other model pages stated their levels in a different sentence shape that this site's scraper couldn't parse. Every current model now publishes its levels; this one's run low, medium, high and xhigh, with no default marked.
It sits in the same rate-limit group as GPT-6 Sol and Astra, so switching costs nothing in throughput. No Batch, Flex or cache-write price is published for it, and no long-context rule is stated on its page — recorded as not published, not assumed to carry over.
The table above is computed directly from this site's facts module: every tier this model publishes, its effort levels, and its rate-limit table.
Verified 2026-10-28 against https://developers.openai.com/api/docs/models/gpt-5.3-codex.
Could not confirm: No Batch or Flex price, no cache-write price, no default effort and no >272K repricing rule are published for this model (checked the pricing page and its model page directly).
Checked: https://developers.openai.com/api/docs/models/gpt-5.3-codex · https://developers.openai.com/api/docs/pricing · https://learn.chatgpt.com/docs/models · https://developers.openai.com/api/docs/deprecations