0.84
- Start
- Sep 15, 202595% CONFIDENCEfrom help.openai.com
GPT-5-Codex Released
OpenAI adds GPT-5-Codex, a GPT-5 variant optimized for agentic coding, across Codex.
- GPT-5-Codex was introduced on September 15, 2025 as a version of GPT-5 optimized for agentic coding; it reached the Responses API on September 23, 2025 but not ChatGPT itself[1][3]
- Launched with a rebuilt Codex CLI (image attachments, to-do list, web search, MCP, three approval modes, compaction), an IDE extension for VS Code, Cursor and forks, and container caching that cut median cloud completion time by 90%[3]
- Code review automatically reviews pull requests moving from draft to ready, or on "@codex review"[3]
- Included in ChatGPT Plus, Pro, Business, Edu and Enterprise plans; recommended only for agentic coding in Codex-like environments[3]
Notable features
- GPT-5-Codex was trained on complex real-world engineering tasks such as building full projects, adding features and tests, debugging, large-scale refactors and code review; OpenAI says it is more steerable, adheres better to AGENTS.md instructions and produces higher-quality code[4]
- It is the default for Codex cloud tasks and code review, and adjusts thinking time to task complexity; it worked independently for more than 7 hours in testing and used 93.7% fewer tokens than GPT-5 on the bottom 10% of turns[3]
- It was trained specifically for code review, navigating the codebase and running code and tests; engineers judged its review comments less likely to be incorrect or unimportant, and in the cloud it can take images or screenshots as input and visually inspect its progress[4]
- In the API (Responses API only) it takes text and image input, with a 400,000-token context window, up to 128,000 output tokens and a September 30, 2024 knowledge cutoff, priced at $1.25 per million input tokens ($0.125 cached) and $10 per million output tokens[2]
- Safety: sandbox with network disabled by default; treated as High capability in Biological and Chemical domains[3]
References 490% CONFIDENCE
The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.
- [1]95%OpenAI Model Release Notes: Introducing GPT-5-codex (Sep 15, 2025)help.openai.com· Posted Sep 19, 2026· Starts Sep 15, 2025· 31% of score
OpenAI's[2][3] own model release notes date this entry Sep 15, 2025
- [2]90%GPT-5-Codex Model | OpenAI APIdevelopers.openai.com· Added Sep 22, 2026· 31% of score
OpenAI's[1][3] API model page for GPT-5-Codex lists Responses-API-only access, a 400,000 context window, 128,000 max output tokens, Sep 30, 2024 knowledge cutoff and $1.25/$10 per 1M tokens.
- [3]85%Introducing upgrades to Codexopenai.com· Added Sep 19, 2026· Starts Sep 15, 2025· 31% of score
OpenAI's[1][2] announcement post, linked from the release note entry.
- [4]85%Introducing upgrades to Codex | OpenAIweb.archive.org· Published Sep 15, 2025· 7% of score
Archived copy of OpenAI's[1][2][3] GPT-5-Codex launch post, which says it is trained on "building full projects from scratch, adding features and tests, debugging, performing large-scale refactors, and conducting code reviews" and "adheres better to AGENTS.md".
Suggest a correction
Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most.
