OpenAI's GPT-5.5 makes computer-use agents the default frontier
Native state-of-the-art computer use, 1M-token context and Codex on Nvidia infra push agentic workflows mainstream
Summary
Openai made GPT-5.5 (and GPT-5.5 Pro) its frontier model on 24 April 2026, building on March's GPT-5.4, the first general-purpose model with native, state-of-the-art computer-use, up to 1M-token context and long-horizon plan/execute/verify. GPT-5.5 powers Codex on Nvidia infrastructure; Codex now serves over 2m weekly users, up 5x in three months, and OpenAI APIs clear 15bn+ tokens/minute. The line collapses dedicated coding models into one agentic model that operates computers and works across spreadsheets, slides and documents. The capability arc tracks rivals: Google's Antigravity agents and Anthropic's Claude Code on Opus 4.8.
By the numbers
- 24 Apr 2026, GPT-5.5 / 5.5 Pro available in the API.
- 1M tokens, context window.
- 2m+, weekly Codex users, up 5x in three months.
- 15bn+, tokens/minute across OpenAI APIs.
Why it matters
Native computer-use moves the frontier from chat to action: an agent that can drive arbitrary software shifts AI from assistant to operator, with direct enterprise-workflow stakes. It also resets evaluation, capability now means task completion across tools, where static coding benchmarks are degrading.
What to watch
- Enterprise deployments of computer-use agents and the error/oversight regime.
- Safety incidents from agents acting across live systems.
- Whether GPT-5.6 ships before the IPO window.