When Machines Learn to Take Their Time
In an era where speed is often prized above all else, OpenAI has taken a different approach with the latest evolution of its coding assistant. The new GPT-5 Codex doesn’t just respond quickly; it responds wisely, even if that takes hours. The upgraded model, announced on September 15, 2025, signals a pivotal shift in how artificial intelligence handles complex programming tasks.
Adaptive Intelligence in Action
At the heart of GPT-5 Codex lies a novel capability: dynamic thinking time. Unlike previous iterations that made quick, static decisions, this new model can assess how much computational effort to dedicate to a task in real-time. Some challenges require only moments; others, GPT-5 Codex may pursue for up to seven hours. This flexibility mirrors human reasoning more than ever before, allowing the model to delve deeper into thorny problems without being hamstrung by arbitrary time constraints.
OpenAI reports that this adaptive capacity allows GPT-5 Codex to outperform its predecessors and competitors on key benchmarks. On SWE-bench Verified, a test measuring how well an AI can autonomously reason through code changes, GPT-5 Codex leads the pack. It also shines in tasks like code refactoring, an often tedious process requiring meticulous logic and contextual understanding.
A Smarter, More Useful Reviewer
Perhaps most notable is the model’s upgraded skill in code reviews. Software engineers testing GPT-5 Codex noted a significant improvement in comment quality: fewer mistakes and more impactful insights. Where earlier AI assistants might overwhelm developers with superficial or incorrect suggestions, GPT-5 Codex offers concise, accurate feedback. This makes it not just a faster assistant, but a genuinely smarter one.
Expanding Access and Integration
The upgraded Codex is now integrated across OpenAI’s product line, including IDEs, terminals, GitHub, and within ChatGPT itself. Users subscribed to the Plus, Pro, Business, Edu, and Enterprise tiers already have access, with broader API rollout expected soon. This strategic deployment ensures the model is available to both individual developers and large teams looking to scale AI-assisted workflows.
Rising Stakes in the AI Coding Race
The launch of GPT-5 Codex comes as competition in the AI coding space intensifies. Players like Claude Code and Anysphere’s Cursor are making headlines of their own, with Cursor reportedly surpassing $500 million in annual recurring revenue this year. For OpenAI, staying ahead means offering not just faster tools, but deeper, more dependable ones. With GPT-5 Codex, they’re betting on quality and longevity over flash.
Questions Still Unanswered
Despite the promise, GPT-5 Codex introduces new complexities. Running models for hours at a time could significantly increase compute costs. Will developers have control over how much “thinking time” the model uses? Will they be able to track progress mid-task? Moreover, while fewer review comments may indicate precision, there’s always the risk that something important might be missed.
OpenAI must also prove that the performance seen in controlled benchmarks holds up across diverse, real-world codebases. From legacy enterprise stacks to cutting-edge startups, the environments are as varied as they are demanding.
The Future of AI Development
OpenAI’s latest move suggests a broader evolution in AI development. Rather than just mimicking human capabilities at superhuman speeds, models like GPT-5 Codex are starting to emulate something more subtle: thoughtful persistence. In the world of software engineering, where complexity is the norm and correctness is critical, that may be the most human trait of all.
Whether GPT-5 Codex lives up to its promise will depend on how it performs outside the lab. But for now, it stands as a compelling vision of what the future of AI-assisted coding could look like, and a strong signal that the days of “fast and shallow” might be giving way to “slow and smart.”
Maya Lindqvist is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Maya Lindqvist's usual length and in Maya Lindqvist's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.