OpenAI improves Codex with GPT-5-Codex and takes it to Chatgpt Plus users and companies

By Canuto

Openai presented GPT-5-Codex, a GPT-5 version designed for its Codex coding agent that dynamically adjusts how much “thinks” of a task. The model is now available for chatgpt plus, PRO, Business, Edu and Enterprise subscribers, and according to TechCrunch improves in coding benchmarks and code reviews supervised by engineers.
***

GPT-5-Codex Adjusts in real time the calculation time for coding tasks; It can take from seconds to 7 hours.
OpenAi deployed it in terminals, IDE, Github and Chatgpt for Plus, Pro, Business, Edu and Enterprise users; The API will arrive later.
According to TechCrunch, GPT-5-Codex exceeds GPT-5 in Swe-Bench Verified and provides code reviews with fewer errors and more impact.

Openai announced a new version of its coding model, called GPT-5-Codex, which is already being deployed in its Codex products as TechCrunch reports.

The company explained that the model dedicates a “thought” time dynamically for each task. That window can vary from a few seconds to 7 hours in some cases, according to the note.

The initial deployment covers accesses from terminals, integrated development environments (IDE), integrations in Github and the chatgpt interface. Access is enabled for chatgpt plus, PRO, Business, Edu and Enterprise users.

🚀⚡ GPT-5-CODEX: OpenAI Upload the bet
Adapts: ‘think’ seconds → 7 hours
Already in Plus/Pro/Business/Edu/Enterprise
Better in Swe-Bench; more precise reviews
API and Costs still without date pic.twitter.com/sedhlliq3W

– Diario ฿ Itcoin (@diariobitcoin) SEPTEMBER 16, 2025

Openai indicated that there are plans to offer GPT-5-Codex to its API customers in the future, without specifying calendar.

What changes about GPT-5

According to TechCrunch, GPT-5-Codex exceeds GPT-5 performance in Benchmarks with agents. In particular, the company cites improvements in Swe-Bench Verified, a benchmark that measures agents-oriented coding skills.

OpenAI also reports advances in code refactorization tests taken from established great repositories. In those trials, GPT-5-Codex obtained better results than GPT-5.

The firm added that the model was specifically trained to make code reviews. Experienced software engineers evaluated the model’s observations and, according to the article, found less incorrect comments and more high -impact comments.

These improvements seek to make Codex more competitive compared to other coding tools with AI in the market.

The dynamics of “thought” and the role of the router

In the technical explanation, Alexander Embiricos, product leader Codex in OpenAI, compared the new GPT-5-Codex dynamics with the GPT-5 router that directs consultations to different models according to its complexity.

Embiricos said that GPT-5-Codex does not use a router under the hood. Instead, the model itself adjusts how long to dedicate to a real -time task.

This difference matters because a router decides at the beginning how much power and time assign. Instead, GPT-5-Codex can begin with an estimate and, minutes later, reallocate resources if it detects that the task requires more calculation.

The Executive added that it has observed cases in which the model came to take more than seven hours to complete a complex task.

Implementation and accessibility

The new model is already integrated into codex products accessible from terminals, IDE, integrations in Github and the Chatgpt experience, according to the source.

OpenAi enabled access for several user categories: Chatgpt Plus, Pro, Business, Edu and Enterprise. No differences between plans in functionality or specific limits were specified.

The company also promised that API customers will be able to access GPT-5-Codex at a later stage, but did not detail dates or API offer prices.

For developers and software equipment, the arrival of GPT-5-Codex implies a tool that adjusts its effort based on the real complexity of the task, and that can offer more relevant code reviews, according to the evaluators cited by Techcrunch.

Competition in the coding market with AI

The launch occurs in a competitive market. Techcrunch lists alternatives such as Claude Code, cursor of Anysphere and Github Copilot de Microsoft.

The article indicates that Cursor exceeded the USD $ 500,000,000 in the early 2025, an indicator of the strong segment growth.

He also mentions the Windsurf case, a code editor that was subject to an acquisition attempt that resulted in team divisions between Google and Cognition.

These references contextualize why Openai seeks to reinforce Codex to sustain positioning against well -financed rivals and growing business adoption.

Implications for developers and companies

For engineers, the promise of more precise and more impact reviews could accelerate workflows and reduce time in manual reviews.

However, it is key that teams validate automatic suggestions. TechCrunch’s article highlights that human engineers evaluated the comments of GPT-5-Codex to measure their quality.

For companies, the possibility that a model dedicates hours to a task poses questions about computational costs and policy of execution time in production.

OpenAI did not detail price structure based on variable computing time or concrete policies on maximum limits, so organizations must wait for more information before planning integrations at scale.

Original image of Diariobitcoin, created with artificial intelligence, for free use, licensed under public domain.

This article was written by an AI content editor and reviewed by a human editor to guarantee quality and precision.

WARNING: Diariobitcoin offers informative and educational content on various topics, including cryptocurrencies, AI, technology and regulations. We do not provide financial advice. Cryptactive investments are high risk and may not be adequate for all. Investigate, consult an expert and verify the applicable legislation before investing. I could lose all its capital.