OpenAI has introduced GPT-5.5which is now available for Plus, Pro, Business and Enterprise users via ChatGPT and Codex. The GPT-5.5 Pro version is also available for Pro, Business and Enterprise subscribers. API access is expected soon, with pricing set at $5 per million entry tokens and $30 per million output tokens.
GPT-5.5 Pro API level will cost $30 per million entry tokens and $180 per million output tokens. OpenAI claims that GPT-5.5 offers similar per-token latency to GPT-5.4 in real-world use, while providing improved performance and requiring fewer tokens to complete the same Codex tasks.
How GPT-5.5 improves agent encoding performance
In Terminal-Bank 2.0which tests complex command-line workflows that require tool planning and coordination, GPT-5.5 achieves 82.7%, compared to 75.1% for GPT-5.4. On Expert-SWE, an internal OpenAI benchmark designed for long-term encoding tasks with an estimated average human completion time of 20 hours, GPT-5.5 also outperforms GPT-5.4, using fewer tokens in all three encoding evaluations.
In SWE-Bench Prothat evaluates real-world GitHub problem solving, GPT-5.5 scores 58.6%. Meanwhile, Claude Opus 4.7 scores higher, at 64.3%.
GPT 5.5 on computer use and knowledge work
In OSWorld-Verifiedwhich evaluates a model’s ability to operate autonomously in real-world computing environments, GPT-5.5 achieves a score of 78.7%, compared to 75.0% for GPT-5.4.
In GDPvalA performance test of agents in 44 different occupations, GPT-5.5 scores 84.9%. In the Tau2-bench Telecom benchmark, which measures performance in complex customer service workflows without quick tuning, GPT-5.5 achieves 98.0%, compared to 92.8% for GPT-5.4.
Scientific research, cybersecurity and security
GPT-5.5 shows gains on GeneBench, a multi-stage scientific data analysis benchmark in genetics and quantitative biology, with a score of 25.0% compared to 19.0% for GPT-5.4. GPT-5.5 Pro scores 33.2% on the same benchmark. On BixBench, a bioinformatics benchmark, GPT-5.5 achieves 80.5% compared to 74.0% for GPT-5.4.
OpenAI also claims that an internal version of GPT-5.5 with a custom harness contributed to a new proof on Ramsey numbers in combinatorics, which was later verified in Lean.
OpenAI rates GPT-5.5’s biological and cybersecurity capabilities as high within its Readiness Framework, just one level below critical. On CyberGym, GPT-5.5 scores 81.8%, compared to 79.0% for GPT-5.4 and 73.1% for Claude Opus 4.7. The update introduces stricter classifiers for potential cyber risks. Additionally, OpenAI is launching a Trusted Access for Cyber program, which allows verified users to request broader access to advanced cybersecurity features with fewer restrictions for defensive purposes.
Where and when can GPT-5.5 be used?
Thought GPT-5.5 is available for Plus, Pro, Business and Enterprise users on ChatGPT. On Codex, GPT-5.5 can be accessed on Plus, Pro, Business, Enterprise, Edu and Go plans with a context window of 400,000 tokens.
A fast mode option generates tokens about 1.5 times faster and about 2.5 times the cost. API access will be available soon. Batch and Flex API pricing is offered at half the standard rate, with priority processing at approximately 2.5 times the standard rate.
