Overview
- OpenAI introduced GPT-5.5 as a self-directed agent for coding and research that can plan tasks, use tools and verify steps.
- GPT-5.5 is available to Plus, Pro, Business and Enterprise users in ChatGPT and Codex, while wider API access is delayed for safety review under a Trusted Access program for sensitive features.
- OpenAI reports strong agentic-coding results, including 82.7% on Terminal-Bench 2.0 and a narrow lead on OSWorld-Verified for desktop control.
- The company’s benchmark tables leave gaps and mixed showings, with missing rival scores on some tests, signs of memorization on SWE-Bench Pro, and rival wins on MCP Atlas tool use and BrowseComp web research.
- API pricing rises to $5 per million input tokens and $30 for output, and OpenAI says lower token use and similar per-token speed could offset the higher rates.