OpenAI just shipped GPT-5.5, seven weeks after GPT-5.4, and OpenAI President Greg Brockman called it “a new class of intelligence.” That is a big claim. It is also useful framing. GPT-5.5 is built for work that crosses tools, screens, files, codebases, spreadsheets, research trails, and humans who occasionally give instructions like, “Just make it work.”
GPT-5.5’s headline numbers look strong. It scores 78.7% on OSWorld-Verified, which measures whether an agent can operate a real computer environment. It scores 84.9% on GDPval, which tests knowledge-work outputs across 44 occupations. In one OpenAI example, a math professor used GPT-5.5 in Codex to build an algebraic-geometry app from a single prompt in 11 minutes. That is an impressive, demo-friendly version of the story.
Then, you get the invoice. OpenAI lists API pricing for GPT-5.5 at $5 per million input tokens and $30 per million output tokens. GPT-5.4 is $2.50 and $15. The rate card doubled, but that doesn’t mean the cost of any given task has doubled.
Model choice is starting to look like cloud architecture. You route workloads. Simple extraction, summarization, translation, classification, and routine analysis go to cheaper models. Gemini Flash-Lite, Gemini Flash, GPT-5.4 mini, and other low-cost models earn their keep here. Code-heavy and long-horizon engineering tasks get tested against the flagship models.
We no longer look at token prices in a vacuum. We calculate cost per completed task. A model that uses fewer retries, fewer human interventions, fewer failed tool calls, and fewer workflow restarts can cost more per token and still be cheaper in production. OpenAI is betting buyers will measure this way. Procurement will try to measure by rate card. Both teams should meet in a conference room with actual workload data and a calculator.
Every company needs a Claw strategy. Do you have one?
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.