OpenAI Unveils GPT-5.5

OpenAI has launched GPT-5.5, positioning it as its most intuitive and intelligent model to date, designed to streamline complex computer-based work. The company claims this iteration understands user intent faster and can autonomously manage tasks, from coding and debugging to data analysis and software operation.

The core advancement lies in GPT-5.5's ability to handle multi-part, ambiguous tasks by planning, utilizing tools, verifying its work, and adapting over time. This is particularly impactful in areas like agentic coding, knowledge work, and scientific research, where sustained reasoning and action are critical.

Despite its increased capabilities, GPT-5.5 reportedly maintains the per-token latency of its predecessor, GPT-5.4, a significant feat for larger models. It also demonstrates improved token efficiency, particularly in coding tasks.

OpenAI emphasizes robust safety measures for this release, including extensive internal and external red-teaming, targeted testing for cybersecurity and biological capabilities, and feedback from early-access partners.

Model Performance and Availability

GPT-5.5 showcases state-of-the-art performance across various benchmarks. On Terminal-Bench 2.0, it achieved 82.7% accuracy in complex command-line workflows. For real-world GitHub issue resolution on SWE-Bench Pro, it reached 58.6%, surpassing previous models in single-pass task completion.

In agentic coding, early testers reported a marked improvement in conceptual clarity and the ability to handle large codebases and ambiguous failures. One tester likened losing access to GPT-5.5 to "having a limb amputated."

For knowledge work, GPT-5.5 excels at information synthesis, data analysis, and document creation. OpenAI itself is leveraging the model for internal workflows, including analyzing speaking request data and processing tax forms more efficiently.

GPT-5.5 Pro, available to Pro, Business, and Enterprise users, reportedly handles more demanding tasks with improved response quality and structure, particularly in fields like business, legal, and data science. It achieved 98.0% on Tau2-bench Telecom, a complex customer-service workflow benchmark.

The model also shows gains in scientific research, with notable improvements on GeneBench for multi-stage genetic data analysis. An internal version even contributed to discovering a new proof in combinatorics.

GPT-5.5 is now rolling out to ChatGPT Plus, Pro, Business, and Enterprise users. API deployments are expected soon.