GPT-5.5 posts 82.7% terminal-bench
- OpenAI said on April 23 it released GPT-5.5, a new model for coding and computer use, and began rolling it out in ChatGPT. - OpenAI’s launch page says GPT-5.5 scored 82.7% on Terminal-Bench 2.0 and matched GPT-5.4 latency while using fewer tokens on Codex tasks. - The API price starts at $5 per million input tokens, signaling a premium model aimed at agent-style work. (openai.com)
OpenAI released GPT-5.5 on April 23 and said the model is built to handle coding, research, data analysis, and software tasks across tools. (openai.com) The company said GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with GPT-5.5 Pro going to Pro, Business, and Enterprise tiers. An April 24 update said GPT-5.5 and GPT-5.5 Pro are now available in the application programming interface, or API. (openai.com) OpenAI’s launch page lists GPT-5.5 at 82.7% on Terminal-Bench 2.0, compared with 75.1% for GPT-5.4, 69.4% for Claude Opus 4.7, and 68.5% for Gemini 3.1 Pro. The same chart lists GPT-5.5 at 78.7% on OSWorld-Verified and 81.8% on CyberGym. (openai.com) Terminal-Bench is a test for terminal-based computer work, the kind of command-line environment developers use to inspect files, run code, and fix systems. OpenAI framed GPT-5.5 as a model that can plan, use tools, check its work, and keep going through multi-step tasks. (openai.com) OpenAI said GPT-5.5 matches GPT-5.4 on per-token latency in real-world serving while performing at a higher level and using fewer tokens on the same Codex tasks. That combination points to a model pitched for longer jobs where speed and cost both matter. (openai.com) The API pricing page lists GPT-5.5 at $5.00 per 1 million input tokens, $0.50 per 1 million cached input tokens, and $30.00 per 1 million output tokens. GPT-5.4 is listed lower, at $2.50 per 1 million input tokens and $15.00 per 1 million output tokens. (openai.com) OpenAI’s system card says the company ran GPT-5.5 through its predeployment safety evaluations and Preparedness Framework, including targeted red-teaming for advanced cybersecurity and biology capabilities. It also said it gathered feedback from nearly 200 early-access partners before release. (openai.com) The public material does not mention a model named “GPT-5.4-Cyber,” and OpenAI’s launch page instead names GPT-5.5, GPT-5.5 Pro, GPT-5.4, and GPT-5.4 Pro. The benchmark claim at the center of the chatter traces back to OpenAI’s own product page, not just reposts on X. (openai.com) The release turns a social-media rumor into a documented product launch: GPT-5.5 is real, the 82.7% Terminal-Bench 2.0 score is on OpenAI’s site, and the listed API entry price is $5 per million input tokens. (openai.com 1) (openai.com 2)
