OpenAI has unveiled GPT-5.5, its latest flagship AI model. The company promises faster reasoning, stronger coding performance, and better ability to complete multi-step tasks across apps and tools. GPT-5.5 understands user intent more quickly and takes on more work independently. Instead of needing step-by-step instructions, the model can plan tasks, use tools, check results, and continue working through unclear requests. With this release, OpenAI launches ChatGPT 5.5 especially strong in software engineering, office work, scientific research, and computer automation.
Faster, Smarter Coding Model
One of the biggest focuses of GPT-5.5 is coding. The model performs better than GPT-5.4 while using fewer tokens, which could lower computing costs and reduce retries. On Terminal-Bench 2.0, a benchmark for complex command-line workflows, GPT-5.5 scored 82.7%, up from 75.1% for GPT-5.4. On OpenAI’s internal Expert-SWE test for long engineering tasks, GPT-5.5 reached 73.1%, compared with 68.5% for GPT-5.4.
| Coding Benchmark | GPT-5.5 | GPT-5.4 |
| Terminal-Bench 2.0 | 82.7% | 75.1% |
| Expert-SWE (Internal) | 73.1% | 68.5% |
The model handles large codebases better, debugs unclear issues, and carries changes across multiple files. OpenAI launches ChatGPT 5.5 as a serious competitor to Claude 4.7.
Better at Workplace Tasks
Beyond coding, GPT-5.5 handles knowledge work such as research, spreadsheet creation, document drafting, and navigating software tools. On GDPval, a benchmark measuring professional task performance across 44 occupations, GPT-5.5 scored 84.9%, ahead of GPT-5.4 at 83.0%. On OSWorld-Verified, which tests whether models can operate computer environments on their own, GPT-5.5 scored 78.7%.
| Productivity Benchmark | GPT-5.5 | GPT-5.4 |
| GDPval | 84.9% | 83.0% |
| OSWorld-Verified | 78.7% | 75.0% |
OpenAI says more than 85% of its own employees already use Codex weekly across teams including engineering, finance, communications, marketing, and product management.
Gains in Research and Math
GPT-5.5 also performs better in scientific workflows that require exploring ideas, analyzing evidence, and testing assumptions over multiple steps. The model scored 25.0% on GeneBench, up from 19.0% for GPT-5.4, and 80.5% on BixBench, compared with 74.0% for GPT-5.4. On FrontierMath Tier 4, GPT-5.5 reached 35.4%, ahead of GPT-5.4 at 27.1%.
Same Speed, Stronger Output
OpenAI launches ChatGPT 5.5 without sacrificing speed. The model matches GPT-5.4 latency while delivering stronger performance. Internal infrastructure improvements increased token generation speeds by more than 20%.
Safety and Availability
GPT-5.5 includes new protections for cybersecurity and biology-related misuse. The API version will launch soon with a 1 million token context window.
