Overview
- OpenAI, which released GPT-5.5 Thursday, began rolling it out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with a higher-accuracy Pro version limited to Pro, Business, and Enterprise and API access coming soon.
- The model takes on messy, multi-part jobs by planning steps, using tools, checking its work, and moving across apps, while matching GPT-5.4 response speed and using fewer tokens on many coding tasks; OpenAI says API pricing will double to $5 per million input tokens and $30 per million output tokens.
- Company benchmarks show gains over GPT-5.4 and some rivals, including 82.7% on Terminal-Bench 2.0 for complex command-line workflows, 78.7% on OSWorld-Verified for operating a computer, and 58.6% on SWE-Bench Pro for one-pass GitHub issue fixes.
- OpenAI says it shipped its strongest safeguards to date, used internal and external red teams, and classified the model’s cyber and biological capabilities as High under its Preparedness Framework, with staged access and a bio bug bounty to curb misuse.
- Early enterprise tests point to practical gains, with the Bank of New York reporting higher accuracy and stronger resistance to hallucinations, which could speed rollout across regulated workflows and shape buying decisions in a tight race with Anthropic and Google.