Introduction#
OpenAI has officially launched its latest AI model, GPT-5.5, which is now accessible to Plus, Pro, Business, and Enterprise users through its ChatGPT and Codex platforms. This new version promises improved capabilities for coding and research applications.
Performance Metrics#
In benchmark tests, GPT-5.5 achieved an impressive 82.7% accuracy on Terminal-Bench 2.0, which assesses command-line workflows. It also scored 58.6% on SWE-Bench Pro, a test focused on resolving issues on GitHub, and 73.1% on Expert-SWE, OpenAI's internal evaluation for coding tasks that typically take around 20 hours to complete.
Efficiency and Development#
OpenAI reports that GPT-5.5 maintains the same speed as its predecessor, GPT-5.4, while using fewer tokens—units of text processed— to accomplish similar tasks. The model was developed using advanced NVIDIA systems, specifically the GB200 and GB300 NVL72.
Specialized Applications#
For professional use, GPT-5.5 scored 84.9% on GDPval, which evaluates knowledge work across various job roles, and 78.7% on OSWorld-Verified, which assesses capabilities in operating computer environments. The model also shows significant improvements in scientific research, achieving 25.0% on GeneBench and 80.5% on BixBench for bioinformatics analysis.
Safety and Access#
OpenAI has rated GPT-5.5’s biological and cybersecurity capabilities as "High" under its Preparedness Framework. Specialized access is available through the "Trusted Access for Cyber" program for verified security professionals. For those interested in using the API, pricing is set at $5 per million input tokens and $30 per million output tokens, with GPT-5.5 Pro costing $30 per million input tokens and $180 per million output tokens. The model underwent thorough safety evaluations, including feedback from around 200 early-access partners before its release.