OpenAI releases GPT-5.5

OpenAI released GPT-5.5 on April 23, saying the new model is built for coding, research, data analysis, and other multi-step work across software tools. (openai.com) OpenAI said GPT-5.5 is rolling out in ChatGPT to Plus, Pro, Business, and Enterprise users, while GPT-5.5 Pro is rolling out to Pro, Business, and Enterprise tiers. The company said API access is coming later, after additional safety and security work for large-scale deployment. (openai.com) The company described GPT-5.5 as a fully retrained model rather than a small update, with gains in writing and debugging code, browsing the web, analyzing files, creating spreadsheets, and operating software until a task is finished. OpenAI said the model matches GPT-5.4 on per-token latency while using fewer tokens on Codex tasks. (openai.com; developers.openai.com) The benchmark numbers OpenAI published are aimed at “agentic” work, meaning jobs where the model plans steps, uses tools, checks results, and keeps going without a human steering every click. OpenAI reported 82.7% on Terminal-Bench 2.0, up from 75.1% for GPT-5.4, and 78.7% on OSWorld-Verified, up from 75.0%. (openai.com) OpenAI also reported 51.7% on FrontierMath Tier 1–3 and 35.4% on Tier 4, alongside 55.6% on Toolathlon and 84.4% on BrowseComp. Those tests measure different pieces of the same problem: solving hard math, choosing and using tools, and completing browser-based tasks with fewer hand-holding prompts. (openai.com) That matters in the part of the AI market that has shifted from chatbots toward software that can carry out office and engineering tasks on a computer. OpenAI’s release says GPT-5.5 is tuned for “complex, real-world work,” and the Codex team says it is now the recommended model for implementation, refactors, debugging, testing, and validation. (openai.com; developers.openai.com) OpenAI has spent the past year moving from general-purpose assistants to models that can act more like junior operators inside apps, browsers, and developer tools. GPT-5 launched in August 2025, GPT-5.4 followed on March 5, 2026, and GPT-5.5 arrives less than two months later with a heavier emphasis on coordination across tools over long tasks. (openai.com; openai.com; openai.com) OpenAI paired the launch with a new system card that says the company ran its full safety and preparedness evaluations, added targeted tests for advanced cybersecurity and biology risks, and worked with internal and external red-teamers. The company said it also collected feedback from nearly 200 early-access partners before release. (openai.com; openai.com) For users, the immediate change is less about a new chat interface than about how much work the model can do before it gets stuck or needs correction. OpenAI is pitching GPT-5.5 as the version that can take a messy assignment, move through tools, and finish more of it on the first pass. (openai.com; openai.com)