GPT-5-Codex
2025-09-15 · Publisher : OpenAI · Domain : Artificial Intelligence
OpenAI releases GPT-5-Codex, a version of GPT-5 optimized for agentic development in Codex, capable of fast interactive sessions as well as long autonomous tasks, with a code review capability.
What changes
- Trained on real-world engineering tasks: full projects, features and tests, debugging, large-scale refactors and code review.
- Adjusts its thinking time to the difficulty: for the shortest 10% of turns, 93.7% fewer tokens than GPT-5; for the longest 10%, about twice as much reasoning time.
- Observed working on its own for more than 7 hours on complex tasks.
- Better adherence to AGENTS.md instructions and higher-quality code outputs, with screenshots as input for front-end work.
- Comes with a revamped Codex CLI and an IDE extension.
### Specifications - Context 400,000 tokens (272,000 input), output 128,000 tokens, text and image input, text output, cutoff September 30, 2024. - Availability at launch: Codex (CLI, IDE, cloud, GitHub, iOS app) with the ChatGPT Plus, Pro, Business, Edu and Enterprise plans. - API: Responses API only, from September 23, 2025, identifier gpt-5-codex. ### Pricing Same price as GPT-5 since the API opened (update to the post dated September 23, 2025). | | Input | Cached input | Output | | --- | --- | --- | --- | | gpt-5-codex | 1.25 | 0.125 | 10 | ### Benchmarks Scores published by the vendor: SWE-bench Verified (500 tasks) and code refactors are shown in charts, with no numerical value in the text. Only the token consumption measurements above are quantified. ### Key takeaways GPT-5-Codex specializes the flagship model for coding agents, with built-in code review. The vendor recommends reserving it for Codex or similar environments, with GPT-5 remaining the general-purpose model.
Links
Tags : LLM, OpenAI, Codex, GPT-5, Code, Agent