With GPT-5.5, GPT-5 is already on its fifth iteration. This model is said to be more intelligent than its predecessor GPT-5.4, while operating at the same speed.
OpenAI introduces GPT-5.5, a new model that, according to the company, performs better in the areas of coding, scientific research, and daily knowledge work. The model combines higher intelligence with the same speed as its predecessor GPT-5.4, and is now available for Plus, Pro, Business, and Enterprise users.
GPT-5.5 is designed to handle complex, multi-part tasks autonomously. While previous models often require step-by-step guidance, GPT-5.5 is reportedly able to plan, deploy tools, and navigate ambiguities until a task is fully completed. According to OpenAI, this makes the model suitable for long-term workflows where human intervention must be kept to a minimum.
Stronger performance in code and knowledge work
In terms of agentic coding, GPT-5.5 scores 82.7% on Terminal-Bench 2.0, compared to 75.1% for GPT-5.4. On Expert-SWE, an internal evaluation for long-term coding tasks with an average execution time of twenty hours for human experts, GPT-5.5 also outperforms its predecessor. Notably, the model achieves these improved results using fewer tokens, making it more efficient to use.
For knowledge work, the model scores 84.9% on GDPval, an evaluation that measures the ability of AI agents to perform specialized tasks across 44 professions. On OSWorld-Verified, which measures whether a model can autonomously operate computer environments, GPT-5.5 achieves a score of 78.7%. More than 85% of employees at OpenAI itself use Codex weekly for a wide range of tasks, from financial analysis to drafting reports.
Availability and pricing
GPT-5.5 is available in ChatGPT for Plus, Pro, Business, and Enterprise users. In Codex, the model is available with a context window of 400,000 tokens. The model will soon be available via the API at five dollars per million input tokens and thirty dollars per million output tokens. GPT-5.5 Pro, intended for the most demanding tasks, costs thirty dollars per million input tokens and 180 dollars per million output tokens.