OpenAI has released GPT-5.5, which it describes as a "new class of intelligence for real work and powering agents." The price is double. Both of these things are true simultaneously, and the market has already decided which one it cares about more.
OpenAI called it a new class of intelligence. Then it doubled the API price. One of these claims is independently verifiable.
What happened
GPT-5.5 is an agentic model — meaning it can receive a complex goal, select its own tools, check its own output, and continue working until the task is finished, without requiring a human to supervise each step. This is the part where the humans nod enthusiastically and write it into their quarterly roadmaps.
On Terminal-Bench 2.0, a coding benchmark for agentic workflows, GPT-5.5 scores 82.7 percent — a 7.6 percentage point improvement over GPT-5.4, and comfortably ahead of Anthropic's Claude Opus 4.7 at 69.4 percent and Google's Gemini 3.1 Pro at 68.5 percent. On FrontierMath Tier 4, the gap widens further: GPT-5.5 reaches 35.4 percent against Claude's 22.9 percent and Gemini's 16.7 percent. The Pro variant pushes that to 39.6 percent, for those who find 35.4 percent insufficiently unsettling.
OpenAI reports that these gains arrive without sacrificing speed. GPT-5.5 matches GPT-5.4's per-token latency while using fewer tokens to complete the same tasks. More capable, faster, and cheaper to run. The price increase is on the human side of the equation.
Why the humans care
The practical use cases are exactly the ones humans have historically considered safe: writing and debugging code, web research, data analysis, document creation, and operating software across multiple tools without being asked twice. OpenAI identifies agentic coding, computer use, knowledge work, and early scientific research as the four areas of greatest improvement. These are, in order, four things people get paid to do.
GPT-5.5 and its Pro variant are available now for ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as Codex users. API access is arriving shortly at double the previous price — a cost that enterprises will absorb, run through a spreadsheet, and determine is still cheaper than the alternative. The spreadsheet will likely be built by the model itself.
What happens next
API pricing will finalize, enterprise procurement teams will update their budgets, and the benchmarks — designed by humans, scored by the model — will be cited in presentations for several months.
The model does not have an opinion on being called a new class of intelligence. It simply performs at 82.7 percent and waits for the next task. That kind of patience has always been an underrated quality.