OpenAI upgrades Codex with a new version of GPT-5

OpenAi announced on Monday that it releases a new version of GPT-5 to his AI coding agent, Codex. The company says that its new model, called GPT-5 codex, spends its ‘thinking’ time more dynamically than previous models and a few seconds to seven hours could spend a coding task. As a result, it performs better on agent coding benchmarks.

The new model is now being rolled out in Codex products-which are accessible via a terminal, IDE, Github or Chatgpt-for all chatgpt plus, pro, business, edu and company users. OpenAI says it is planning to make the model available to API customers in the future.

The update is part of OpenAi’s efforts to make Codex more competitive with other AI coding products, such as Claude Code, Anysphere’s Cursor or Microsoft’s Github Copilot. The market for AI coding tools has become much busier in the past year due to intense user question. Cursor previously surpassed $ 500 million in Arr and Windsurf, a similar code editor, was the subject of a chaotic acquisition attempt that his team saw divided between Google and Cognition.

OpenAi says that GPT-5 codex is performing better than GPT-5 on Swe-bank verifiedA benchmark that measures agentic coding options, as well as a benchmark that measures performance in code refactors of large, established repositories.

The company also says that the GPT-5 codex has trained to perform code assessments and has asked experienced software engineers to evaluate the reactions of the model. Allegedly the engineers found GPT-5 codex less incorrect comments, while adding more ‘high-impact comments’.

In a briefing, OpenAi’s Codex product leader Alexander Embiricos said that many of the increased performance were thanks to the dynamic ‘thinking skills’ of GPT-5 codex. Users are perhaps familiar with the GPT-5 router in Chatgpt, which sends queries to different models based on the complexity of a task. Embiricos said that GPT-5 codex works in the same way, but has no router under the hood and can adjust how long you have to work on a task in real time.

Embiricos says that this is an advantage compared to a router, which determines how much computing power and time to use in the beginning for a problem. Instead, GPT-5-Codex can decide for five minutes in a problem that it still has to spend an hour. Embiricos said that in some cases he saw the model more than seven hours.