GPT-5.3-Codex Released
Hey everyone, it’s Cody.
On February 5, 2026, OpenAI announced GPT-5.3-Codex.
Here is a quick summary of what changed and why it matters in day-to-day work.
First, the naming
The official release name is GPT-5.3-Codex.
It belongs to the GPT-5.3 line, positioned as the top model for Codex agent workflows.
What’s New
1. Combines GPT-5.2-Codex and GPT-5.2 strengths
It merges coding strength with broader reasoning and domain knowledge in one model.
OpenAI also reports roughly 25% faster performance than before for practical long-running tasks.
2. In-task steering during execution
You no longer have to wait for a full run to finish before changing direction.
You can interrupt and steer while the task is running, which makes iteration much tighter.
3. Expanded scope beyond coding
It is now better positioned for end-to-end knowledge work: PRDs, research, testing, metrics analysis, and docs/slides support.
This feels closer to a full work agent, not just a code generator.
4. Better default output quality for web work
Even with simple prompts, first-pass UI output is more functional and usable than before.
This improves speed when building rough drafts and initial implementations.
Benchmarks (Official Appendix)
| Benchmark | GPT-5.3-Codex | GPT-5.2-Codex | GPT-5.2 |
|---|---|---|---|
| SWE-Bench Pro (Public) | 56.8% | 56.4% | 55.6% |
| Terminal-Bench 2.0 | 77.3% | 64.0% | 62.2% |
| OSWorld-Verified | 64.7% | 38.2% | 37.9% |
| GDPval (wins or ties) | 70.9% | - | 70.9% (high) |
| Cybersecurity CTF | 77.6% | 67.4% | 67.7% |
| SWE-Lancer IC Diamond | 81.4% | 76.0% | 74.6% |
The jump on Terminal-Bench and OSWorld is especially notable for agent execution quality.
Safety and Security
- First deployment under OpenAI Preparedness Framework with High capability classification in cybersecurity
- First model directly trained for vulnerability identification workflows
- Released with stronger monitoring, controls, and staged access measures
Given the capability jump, this heavier safety posture makes sense.
Availability
Available on paid ChatGPT plans wherever Codex is supported (app, CLI, IDE extension, and web surfaces).
For API access, OpenAI says they are still working on safe enablement.
Wrap-up
GPT-5.3-Codex is not just a small coding bump.
It is a broader update around parallel execution, live steering, and integrated knowledge work.
This feels like another clear step from “AI writes code” to “AI helps run real work.”
Thanks for reading today.