OpenAI has released a new model, GPT-5.5, which the company calls “a new class of intelligence for real-world work.” The main focus of the update is on “agency” – the neural network’s ability to not just answer questions, but to independently perform complex computer tasks, plan its actions, and verify the results without human intervention.
The neural network performs better in Codex programming (finding errors and refactoring), scientific research, and office work. Thanks to deep integration with Nvidia GB200 systems, the model runs faster and uses tokens more efficiently.
The benchmark results were as follows:
| GPT-5.5 | GPT-5.4 | Claude Opus 4.7 | Gemini 3.1 Pro | |
| GDPval (skills in 44 professions) | 84.9% | 83.0% | 80.3% | 67.3% |
| Terminal-Bench 2.0 (code and console) | 82.7% | 75.1% | 69.4% | 68.5% |
| OSWorld-Verified (working with OS) | 78.7% | 75.0% | 78.0% | — |
| Expert-SWE (complex programming) | 73.1% | 68.5% | — | — |
The model and its Pro version are already available to users of paid plans (Plus, Pro, Business, and Enterprise) in ChatGPT and Codex. API access will be available soon.
At the same time, the price for developers has increased significantly – the basic GPT-5.5 will now cost $5 per million tokens for input and $30 for output. GPT-5.5 Pro will also be available in the API, priced at $30 and $180 per million tokens, respectively.
Source:
Image courtesy of OpenAI.
