OpenAI releases GPT-5.5: stronger at coding, agents and computer use

OpenAI has introduced GPT-5.5, a new base model with a clear focus on programming, agentic scenarios and computer use. According to the company, this is not a cosmetic update but a noticeable step forward in tasks where a model has to do more than answer: it has to act more independently.

Where the model improved

GPT-5.5 centers on practical tasks where code accuracy, stability across long chains of actions and confident work with interfaces matter. OpenAI says the model writes better code, handles computer use better and can complete more autonomous tasks without constant human oversight.

For users, this means fewer manual corrections in typical work scenarios, from writing functions to completing multi-step processes.

OpenAI also stresses that GPT-5.5 is the first fully retrained base model since GPT-4.5. That is an important signal: this is not just tweaked settings on top of an old foundation, but a deeper rework of the model itself. Gains in agentic coding tests indirectly support this. Such benchmarks show how well a model holds context, plans steps and sees a task through, rather than just generating plausible text.

Where it is available

The release is already rolling out to ChatGPT and Codex users, but access depends on the plan. Base GPT-5.5 has appeared in the Plus, Pro, Business and Enterprise plans. For those who need maximum performance, OpenAI is separately preparing GPT-5.5 Pro, intended for Pro, Business and Enterprise. In other words, the company splits the mainstream and the more expensive professional use cases from the start. That makes the launch notable not only for enthusiasts but also for corporate teams.

In practice, it looks like this:

• GPT-5.5 is already available in ChatGPT
• The model has also appeared in Codex
• Plus, Pro, Business and Enterprise plans are supported
• GPT-5.5 Pro is a separate version
• API access for developers is still marked as coming soon

This launch order reflects OpenAI's familiar strategy: first give the model to end users inside its own products, then open it through the API to outside teams and services. For developers, this means a pause: they cannot test the new model in production integrations right now, even though the release is already drawing interest in coding and agentic scenarios. Until then, the model can only be compared inside OpenAI's interfaces.

Price and expectations

The most contentious point is cost. The per-token price of GPT-5.5 is twice that of GPT-5.4. For companies that run large volumes of requests, this is not a cosmetic difference but a direct recalculation of product economics. If the new model is genuinely and noticeably better at programming and autonomous work, part of the market will accept the increase. But if the gain turns out to be smaller in real tasks, the price debate will quickly move to the front.

This is the first fully retrained base model since GPT-4.5.

This wording sets the main tone of the release. GPT-5.5 is presented as an update that should be felt in daily work, not just in the name. That is why attention will be especially high among developers, teams building agentic products and companies using models in partly autonomous processes. They will be the first to check whether the promises of better code, computer use and autonomy hold up outside demos and tests.

For now, the picture is mixed. On one hand, the release looks stronger than an ordinary minor update and may become an important milestone after GPT-5.4. On the other, a full assessment will begin only once the API arrives and the first independent comparisons of price, quality and stability in real pipelines appear. That stage will show whether this is a real shift or just an expensive improvement on paper.

What this means

GPT-5.5 shows that OpenAI is again shifting its focus from chat for chat's sake to models that must deliver measurable value in code, agents and workflows. If the quality gain is confirmed in practice, the market gets not just a new version but a more suitable tool for autonomous AI scenarios, albeit at a noticeably higher price. For product teams, this is a reason to recalculate where a more expensive model pays off.