OpenAI Introduces GPT-5.5 for Advanced Coding, Research, and Cybersecurity

OpenAI announced in a press release the launch of GPT-5.5, its most capable model yet for professional and research tasks. The model is designed to handle complex coding, data analysis, document creation, and computer operations while maintaining the same latency as GPT-5.4. GPT-5.5 is available in ChatGPT and Codex for Plus, Pro, Business, and Enterprise users, with API access planned soon.

GPT-5.5 shows significant gains in coding benchmarks, achieving 82.7% on Terminal-Bench 2.0 and 58.6% on SWE-Bench Pro. It uses fewer tokens to complete tasks, improving both efficiency and accuracy. The model also demonstrates improved reasoning and persistence in long coding workflows, outperforming GPT-5.4 and competing models such as Claude Opus 4.7 and Gemini 3.1 Pro.

In knowledge work, GPT-5.5 scored 84.9% on GDPval and 78.7% on OSWorld-Verified, indicating stronger performance in document generation, spreadsheet modeling, and tool use. Scientific evaluations showed improvements in tasks like bioinformatics and genetics analysis, with 80.5% on BixBench and 25% on GeneBench. The model also participated in mathematical research, contributing a verified proof in combinatorics.

GPT-5.5 introduces enhanced cybersecurity safeguards and supports a Trusted Access for Cyber program to expand secure use in critical infrastructure defense. The model was co-designed with NVIDIA GB200 and GB300 NVL72 systems for efficient inference, achieving faster token generation through improved load balancing. GPT-5.5 Pro targets high-accuracy enterprise work, with pricing starting at $5 per million input tokens and $30 per million output tokens for API use.