OpenAI has announced the launch of the new language model GPT-5.5, highlighting a significant increase in its capabilities in programming, scientific research, and working with various tools.
This is reported by Business • Media
GPT-5.5: Enhanced Capabilities and a New Approach to Integration
GPT-5.5 is positioned as the most intuitive and “smart” model in OpenAI’s product line. According to the developers, the new model better understands user intentions and requires fewer tokens to perform tasks, which increases efficiency when working with it. The model is capable of executing a wide range of tasks—from writing and optimizing code to data analysis, document preparation, and switching between different tools. This allows it to take on more functions that are typically performed by a human at a computer.
“We are moving towards more agent-like and intuitively understandable computing.”
According to OpenAI President Greg Brockman, this release is an important step towards creating a “super app”—a universal service that will integrate ChatGPT, Codex, and AI Browser. The company expects that such integration will allow users to perform most work tasks within a single ecosystem.
Focus on Programming, Science, and Security
GPT-5.5 demonstrates significant improvements in programming: in the Terminal-Bench 2.0 tests, it scored 82.7%, and in SWE-Bench Pro—58.6%. In the internal benchmark Expert-SWE for complex engineering tasks, the model outperformed the previous version GPT-5.4 while reducing token consumption.
In the GDPval tests, which assess intellectual work across 44 professions, GPT-5.5 scored 84.9%. In the OSWorld-Verified benchmark, which tests performance in real computing environments, it scored 78.7%. During the Tau2-bench Telecom trials, which involve complex customer service scenarios, the model achieved 98.0% without additional tuning. High results were also noted in financial analysis, modeling, and solving office tasks.
OpenAI also focused on scientific research: in GeneBench, a test for genetics and quantitative biology, GPT-5.5 significantly outperformed its predecessor. In BixBench for bioinformatics, the model achieved the best result among all systems with published scores. According to the company, GPT-5.5 can already accelerate processes in real scientific work.
Significant attention was paid to security: for GPT-5.5, control over dangerous queries has been strengthened, new risk classifiers have been added, and measures against misuse have been expanded. The model’s capabilities in cybersecurity, as well as in biological and chemical fields, are assessed by the company as “high level.”
GPT-5.5 is already available to users on Plus, Pro, Business, and Enterprise plans in ChatGPT and Codex, while the GPT-5.5 Pro version is available for Pro, Business, and Enterprise. The API for the model is expected to be opened soon. The cost of the basic version will be $5 for 1 million input tokens and $30 for 1 million output tokens, while the Pro version will be $30 and $180 respectively.