The initial rollout covered ChatGPT and Codex; an April 24 update added availability in the API.
What happened
OpenAI reports gains in coding, computer use, knowledge work and scientific tasks. It says the model matches GPT-5.4's per-token latency while using fewer tokens on comparable Codex tasks.
Why it matters
An agent working across applications must plan, choose tools and check progress as conditions change. Efficiency can matter as much as a benchmark gain when a task takes many sequential steps.
The bigger picture
Our take. Per-token speed is only part of the user experience. Compare total completion time, tool delays and review effort, and treat the provider's results as a starting point for checking the tasks that matter to you.
Source checked November 18, 2026. Company claims are attributed; paragraphs marked “Our take” are editorial interpretation.