OpenAI on Thursday released its most capable AI system, GPT-5.5, which the company says will enable a more powerful Codex coding agent. OpenAI is quick to say, however, that GPT-5.5 will power the widening set of general digital work tasks that Codex is capable of. The system is significantly better than previous releases at helping with scientific work, including creative aspects of generating new hypotheses and testing them.
The system represents an improvement in autonomous or agentic capability. GPT-5.56 “represents a step toward AI systems that can complete complex, multi-step tasks on a computer without human guidance,” OpenAI says in a blog post.
GPT-5.5, OpenAI says, scores better than any other AI system on the Terminal-Bench 2.0 benchmark, which tests for complex command-line workflows requiring planning, iteration, and tool coordination. It scores 82.7%, compared to 75.1% for GPT-5.4, 69.4% for Anthropic’s Opus 4.7, and 68.5% for Google’s Gemini 3.1 Pro. On OSWorld-Verified, which measures whether a model can operate a computer independently, it scores 78.7% versus 75.0% for GPT-5.4. Anthropic’s Opus 4.7 scored 78%.
Anthropic’s newest frontier model, called Mythos, is finished but has not been publicly released.
OpenAI says it’s seen a surge in users for its Codex coding agent, with about 4 million developers using the tool weekly. During a call with press, OpenAI CEO Greg Brockman said GPT-5.5 will enable Codex to produce polished code, and go about coding projects with the judgement of a senior software engineer.
OpenAI is positioning GPT-5.5 as its strongest agentic coding model. On SWE-Bench Pro, a benchmark measuring real-world GitHub issue resolution, it resolves 58.6% of tasks end-to-end in a single pass. Developers who tested the system early said GPT-5.5 has a better understanding of the “shape” of a software system, and can better understand why something is failing, where the fix is needed, and what else in the codebase would be affected.
The GPT-5.5 release comes just weeks after the release of the GPT-5.4 system. OpenAI is under pressure to keep pace with its rival Anthropic–especially in the area of AI coding assistants, which have proved to be AI’s biggest impact on business operations to date. The pace of system releases is accelerating because the AI itself is doing a lot of the heavy lifting of coding the AI systems themselves.
Some have speculated that the new OpenAI model might be as large as 10 trillion parameters, but Brockman declined to say.
GPT-5.5 is rolling out to OpenAI’s Plus, Pro, Business, and Enterprise subscription users in ChatGPT and Codex. GPT-5.5 Pro, a higher-accuracy version, is available to Pro, Business, and Enterprise users.
