OpenAI unveiled a new artificial intelligence (AI) model, GPT-5.5, two months after its last release.
According to OpenAI on the 24th, the newly released GPT-5.5 is designed to support real-world work across coding, online research, data analysis, document and spreadsheet creation, and software operation.
Improvements were also confirmed in performance metrics. In GDPval, which assesses knowledge-work capability, it scored 84.9%, surpassing rival Anthropic's Opus 4.7 (80.3%).
In Terminal-Bench 2.0, which measures complex command-line task performance, it scored 82.7%, beating both GPT-5.4 (75.1%) and Opus 4.7 (69.4%). In CyberGym, which evaluates cybersecurity capability, it also posted 81.8%, higher than Opus 4.7 (73.1%).
In SWE-Bench Pro, which evaluates problem-solving ability in environments similar to actual development settings, it scored 58.6%, showing improvement over the previous model. The share of tasks solved in a single pass increased, boosting efficiency across the entire development process, including coding, debugging, testing, and verification, it said. However, it fell short of Opus 4.7 (64.3%).
Safety was also strengthened. GPT-5.5 applies more robust safeguards and introduces stricter refusal standards and additional protections for high-risk cyber-related requests.
GPT-5.5 will be rolled out sequentially starting today to ChatGPT Plus, Pro, Business, and Enterprise users. The high-performance version, GPT-5.5 Pro, will be available to Pro, Business, and Enterprise users. The API will be released separately at a later date.