- GPT-5-CODEX promises higher performance and success rates
- It is included for more users, Pro, Business, Edu and Enterprise
- The model can use 93.7% less tokens on light tasks
OPENAI has shared more details on GPT-5-CODEX, a specially designed version of GPT-5 specifically optimized for agency coding and software engineering from the real world, and we are for a treat in terms of reliability and performance.
The chatgpt manufacturer said that a 74.5% reference success rate checking Swee-Bench, the refactoring performance improved at 51.3% (against 33.9% in GPT-5).
Like GPT-5, GPT-5-CODEX will dynamically adjust the reasoning time for faster performance on small tasks and a more complete reasoning on complex performance, and it has already been tested by working independently for more than seven hours on large refectors.
GPT-5-CODEX is a large upgrade
OPENAI says GPT-5-CODEX is solid in code criticisms, catching critical bugs before publication, but it can also manage the work of the front with visual inspection, screenshots and mobile web design improvements.
The news comes only a few months after Openai launched Codex CLI (April) and Codex Web (in May), before combining them in a “unified experience … connected by … Chatgpt” in early September.
It is included with Chatgpt Plus, Pro, Business, Edu and Enterprise Plans, and works on terminal, IDE, on the web, in Github and on the iOS application.
The company also detailed how GPT-5-CODEX uses 93.7% less tokens than GPT-5 on light interactions, but it will also pass twice as much reasoning, editing, tests and iteration if it needs it.
Equally important for developers, the tool will provide newspapers, quotes and test results for transparency. Developers using Codex CLA via an API key will also have access to the API to the “soon” GPT-5 codex.
“Codex becomes the coding partner that we have always considered – the one that is faster, more reliable and deeply integrated into the tools you already use,” wrote Openai.
In addition, the Edu and Business plans are enough to cover “a few coding sessions targeted each week” – users who need more should go to Pro for “a full week of work on several projects”. Company accounts pay what they use via a shared credit pool.
