File photo: Openai has upgraded its AI coding agent, Codex, to its latest GPT-5 AI model, CEO Sam Altman announced on X. | Photo credit: Reuters
Openai has upgraded its AI coding agent, Codex, to its latest GPT-5 AI model, CEO Sam Altman announced to X. Called GPT-5 Codex, the agent can choose how much time it spends on a task, with more complex problems taking up to 7 hours at a time.
The firm said that GPT-5-CODEX outperforms GPT-5 on coding tasks, outperforming it on benchmarks such as SWECH, which measures the coding abilities of agents, as well as others that test code reformatting tasks from large repositories.
A blog post by Openai stated that GPT-5 Codex used 93.7% fewer tokens than GPT-5, while being able to spend twice as much time reasoning, editing, testing, and coding during iterations.
GPT-5 Codex is also trained to review code and more accurately look for errors with “high-impact comments.”
The market for AI coding tools has become very competitive with competitors like Claude Code, Cursor, and Replit coming to the fore.
The updated version is being rolled out to Codex for all Chatgpt Plus, Pro, Business, EDU, and Enterprise users. Openai will also make it available via API at a later date.
Published – September 16, 2025 at 11:56 AM IST
