Highlights
- OpenAI has launched GPT-5.5, its newest model, and says it performs better on coding, computer use, office work, and early research tasks.
- The model is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with API access coming soon.
- OpenAI says GPT-5.5 is more persistent on long tasks and more token-efficient than earlier versions.
- The company points to gains on benchmarks like GDPval, OSWorld-Verified, and Tau2-bench Telecom.
- Early testers say GPT-5.5 handles multi-step work with less hand-holding and better tool use.
Key Facts
| Topic | Details |
| New model | GPT-5.5 |
| Main strengths | Coding, computer use, office tasks, research |
| ChatGPT access | Plus, Pro, Business, Enterprise |
| Codex access | Plus, Pro, Business, Enterprise, Edu, Go |
| API status | Coming soon |
| Noted benchmarks | GDPval 84.9%, OSWorld-Verified 78.7%, Tau2-bench Telecom 98.0% |
If a business team needs an AI model that can stay with a task longer, follow instructions more reliably, and move through messy work with less back-and-forth, GPT-5.5 is OpenAI’s latest answer. The company says the new model is built for coding, research, document-heavy work, and computer use, with stronger performance across long, multi-step tasks.
OpenAI says GPT-5.5 is now rolling out in ChatGPT and Codex for multiple paid tiers, while API access will follow soon. It also says the model is more token-efficient than GPT-5.4, which matters for users who run large workloads or spend a lot of time inside the product.
Background
OpenAI introduced GPT-5.5 on April 23, 2026, and described it as a major step forward for practical work. Axios reported that the company sees it as a move toward more “agentic” and intuitive computing, where the model can handle longer tasks with less prompting.
On OpenAI’s own release page, the company says GPT-5.5 does especially well in settings that mirror real work. That includes office tasks, coding, analysis, and computer use. OpenAI highlighted benchmark scores of 84.9% on GDPval, 78.7% on OSWorld-Verified, and 98.0% on Tau2-bench Telecom.
The company also says GPT-5.5 performs better on scientific and technical research tasks. OpenAI pointed to progress on GeneBench and BixBench, and said an internal version even helped discover a new proof about Ramsey numbers before it was verified in Lean.
For everyday users, the main takeaway is simple: GPT-5.5 is meant to feel more useful in real workflows. It is designed to keep going on complex jobs, use tools more cleanly, and give answers that are more structured and more reliable. OpenAI says GPT-5.5 Pro is also available for harder questions and higher-accuracy work.
FAQ
What is GPT-5.5?
GPT-5.5 is OpenAI’s latest model, built for coding, research, office tasks, and computer use.
Who can use GPT-5.5 right now?
OpenAI says it is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with wider API access coming soon.
What makes GPT-5.5 different from earlier models?
OpenAI says it is more persistent, more token-efficient, and stronger at long, multi-step work than GPT-5.4.
Does GPT-5.5 help with research work?
Yes. OpenAI says it performs better on scientific and technical research workflows, including data analysis and proof discovery.
Is GPT-5.5 available in the API yet?
OpenAI says API access will arrive soon.
