GPT-5.2-Codex
2025-12-18 · Publisher : OpenAI · Domain : Artificial Intelligence
OpenAI releases GPT-5.2-Codex, its most advanced version of GPT-5.2 for agentic development, better at large refactors and migrations, on Windows and in cybersecurity.
What changes
- Improved long context and native compaction for long-running tasks, refactors and code migrations.
- State of the art announced on SWE-Bench Pro and Terminal-Bench 2.0 (values in charts).
- More reliable in native Windows environments and better interpretation of screenshots, diagrams and mockups.
- Cybersecurity capabilities superior to all previous models, without reaching the “High” level of OpenAI's preparedness framework; invitation-only trusted access program for defenders.
- A researcher found React flaws related to React2Shell with GPT-5.1-Codex-Max, according to the vendor.
### Specifications - Context 400,000 tokens (272,000 input), output 128,000 tokens, text and image input, text output, knowledge cutoff August 31, 2025. - Reasoning efforts low, medium, high and xhigh; identifier gpt-5.2-codex. - Available on the day of the announcement in all Codex surfaces for paid subscribers; API announced “in the coming weeks”. ### Pricing Pricing not disclosed at launch. The current documentation indicates the same prices as GPT-5.2. | | Input | Cached input | Output | | --- | --- | --- | --- | | gpt-5.2-codex (documentation) | 1.75 | 0.175 | 14 | ### Benchmarks No numerical value in the blog post text; the results are presented as charts. ### Key takeaways GPT-5.2-Codex is aimed at teams that delegate heavy refactors to an agent. Its offensive and defensive security capabilities justify a supervised deployment.
Links
Tags : LLM, OpenAI, Codex, GPT-5.2, Code, Cybersécurité