OpenAI launches GPT-5.5

OpenAI launches GPT-5.5, a powerful model achieving 82.7% on Terminal-Bench 2.0, now available for ChatGPT Plus, Pro, Business, and Enterprise subscribers.

OpenAI announces GPT-5.5, presented as its most powerful model to date, with measured gains particularly in agentic coding, autonomous computer use, and scientific research. On Terminal-Bench 2.0, which evaluates complex command-line workflows, the model achieves 82.7% compared to 75.1% for GPT-5.4. On OSWorld-Verified, which measures the ability to autonomously operate real computing environments, it reaches 78.7%. OpenAI also highlights improved token efficiency: GPT-5.5 reportedly produces superior results while consuming fewer tokens than its predecessor on the same coding tasks. A Pro version is available for the most demanding uses. Scientifically, an internal version of the model contributed to establishing a mathematical proof concerning Ramsey numbers, subsequently verified in the Lean proof assistant. Its cybersecurity capabilities are rated "High" within OpenAI's internal evaluation framework, without reaching the "Critical" threshold; enhanced safeguards accompany the deployment, with expanded access reserved for verified defense actors. GPT-5.5 is available starting today for Plus, Pro, Business, and Enterprise subscribers in ChatGPT and Codex. API access is announced for the coming weeks, priced at 5 dollars per million input tokens and 30 dollars per million output tokens. Flavio Adamo, a developer and daily user of language models, published a detailed feedback report on 𝕏 regarding GPT-5.5 after several weeks of intensive use. His conclusion: GPT-5.5 does not represent a spectacular leap, but it sufficiently reduces usage friction to make returning to GPT-5.4 difficult.