openai.com
|
ksl
|
|
OpenAI released GPT-5.4 in three tiers – standard, Thinking and Pro – with a one-million-token context window on the API and native computer use baked in for the first time. The model scores 33% fewer factual errors per claim than GPT-5.2 and set records on OSWorld-Verified and WebArena Verified, both computer use benchmarks. That computer use piece is the real story. Anthropic shipped it with Claude months ago, and Google has been demoing similar capabilities with Gemini, but OpenAI positioning it inside a general-purpose model rather than a separate agent product changes the competitive framing. Pro access stays locked behind the $200 ChatGPT Pro tier, while Thinking rolls out to all paid subscribers.
