OpenAI released GPT-5.5 on Thursday, calling the model its "smartest and most intuitive" to date and positioning it as a significant step toward building a unified platform encompassing ChatGPT, Codex, and browser capabilities. Recently, OpenAI has been focused on simplifying and refining its product lineup and prioritizing its enterprise offerings. More specifically, the company has shared its plans to build a "super app" that combines the best features of its current products. This plan seems to have begun taking shape with a recent update to Codex, which, among other things, makes it possible for Codex to use other computer apps in the background, alongside the user, and to explore the web using its in-app browser.

According to OpenAI's announcement, GPT-5.5 delivers substantial improvements across agentic coding, computer use, knowledge work, and scientific research while maintaining GPT-5.4's per-token latency. Notably, GPT-5.5 achieves state-of-the-art performance while using fewer tokens to complete tasks, making it both more capable and more efficient. Performance improvements are, as expected, reported in terms of benchmark scores. Notably, GPT-5.5 scored 82.7% on Terminal-Bench 2.0 (complex command-line workflows), 58.6% on SWE-Bench Pro (GitHub issue resolution), and 84.9% on GDPval (knowledge work tasks). These scores put GPT-5.5 ahead of competitors including Claude Opus 4.7 and Gemini 3.1 Pro.

"This model is a real step forward towards the kind of computing that we expect in the future," said OpenAI president Greg Brockman, adding that it represents progress toward more "agentic and intuitive computing."

GPT-5.5's research capabilities were measured against GPT-5.4 in certain benchmarks. In particular, GPT-5.5 achieved a stronger performance than its predecessor on GeneBench for genetics analysis and BixBench for bioinformatics. OpenAI claims that GPT-5.5 is now capable of providing meaningful scientific, especially in mathematical and biomedical research (the company goes as far as calling the model a "bona-fide co-scientist"). Perhaps the strongest piece of evidence offered to support this claim is the fact that an internal version of the model seems to have helped researchers find a new proof about Ramsey numbers, which are mathematical objects studied by a specific branch of combinatorics.

OpenAI says it has deployed enhanced cybersecurity safeguards for GPT-5.5, having rated it as having "high" cybersecurity, biological and chemical capabilities, as specified under the company's Preparedness Framework. The release includes stricter controls around higher-risk cyber activities while expanding defensive capabilities through a new "Trusted Access for Cyber" program. This program will enable a very restricted group of verified customers to access more permissive models for legitimate applications.

GPT-5.5 is now available to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with API access coming soon at $5 per million input tokens and $30 per million output tokens.