OpenAI has launched GPT-5.5, its most advanced model yet for coding, research and computer-based work.
The new model is rolling out in ChatGPT and Codex, with stronger reasoning, faster task completion and wider agentic capabilities.
A Bigger Push Into Agentic Work
OpenAI GPT-5.5 is designed for users who want AI to complete multi-step tasks instead of only answering questions.
The model can write and debug code, research online, analyze data, create documents, build spreadsheets and operate across tools with less step-by-step guidance. OpenAI says GPT-5.5 can handle messy tasks by planning, using tools, checking results and continuing through uncertainty.
That makes this release important beyond software development. OpenAI is positioning GPT-5.5 as a model for everyday knowledge work, where people need help turning raw information into useful output.
Codex Gets a Major Upgrade
The biggest early impact appears inside Codex, OpenAI’s agentic coding app.
GPT-5.5 scored 82.7% on Terminal-Bench 2.0, ahead of GPT-5.4’s 75.1%. It also reached 58.6% on SWE-Bench Pro and 73.1% on OpenAI’s internal Expert-SWE benchmark.
In practical terms, OpenAI says GPT-5.5 is better at understanding large codebases, debugging failures, testing assumptions and carrying changes across related files.
NVIDIA employees who tested GPT-5.5-powered Codex reported faster development cycles, with debugging work moving from days to hours and larger experiments progressing overnight.
Not Just Coding: Research, Office Work and Science
OpenAI is also highlighting GPT-5.5 as a stronger model for business and research tasks.
The model scored 84.9% on GDPval, a benchmark focused on professional knowledge work, and 78.7% on OSWorld-Verified, which tests whether a model can operate real computer environments.
OpenAI also says GPT-5.5 performs better in scientific workflows, including genetics, bioinformatics and mathematical reasoning. In one example, an internal GPT-5.5 system helped discover a mathematical proof related to Ramsey numbers, later verified in Lean.
That is unusual because it pushes the model beyond simple automation and closer to acting like a research collaborator.
NVIDIA Hardware Powers the Release
GPT-5.5 was co-designed, trained and served on NVIDIA GB200 and GB300 NVL72 systems, according to OpenAI. The company says this helped keep GPT-5.5 at GPT-5.4-level latency while delivering higher intelligence.
NVIDIA has also deployed GPT-5.5-powered Codex internally across teams, including engineering, product, finance, legal, marketing and operations.
The partnership is not new. NVIDIA and OpenAI have worked together for years, with NVIDIA hardware forming a major part of OpenAI’s training and inference infrastructure.
Stronger Safety Controls for Cybersecurity
OpenAI says GPT-5.5 is being released with tighter safeguards, especially around cybersecurity and biology-related capabilities.
The company says it tested the model through internal and external red-teaming, added targeted checks for advanced cybersecurity and biology risks, and collected feedback from nearly 200 early-access partners before launch.
GPT-5.5 is being treated as “High” for biological/chemical and cybersecurity capability under OpenAI’s preparedness framework, but OpenAI says it did not reach the “Critical” cybersecurity level.
Availability and API Plans
GPT-5.5 is rolling out to Plus, Pro, Business and Enterprise users in ChatGPT and Codex. GPT-5.5 Pro is available for Pro, Business and Enterprise users in ChatGPT.
For Codex, GPT-5.5 is available to Plus, Pro, Business, Enterprise, Edu and Go plans with a 400K context window.
OpenAI says GPT-5.5 and GPT-5.5 Pro will come to the API soon. The planned API pricing is $5 per 1 million input tokens and $30 per 1 million output tokens for GPT-5.5, while GPT-5.5 Pro is planned at $30 per 1 million input tokens and $180 per 1 million output tokens.
The Bigger Shift
GPT-5.5 shows where AI products are moving next.
The race is no longer only about answering prompts. OpenAI is trying to make AI act more like a computer-based worker that can plan, execute, review and finish complex tasks.
For users, that could mean faster coding, deeper research, cleaner documents and more useful automation. For companies, it signals a bigger change: AI agents are becoming part of daily work, not just experimental tools.