AI Daily Brief — 5 March 2026

OpenAI ships GPT-5.4 — the first general-purpose model with native, state-of-the-art computer-use capabilities. The headline number: 75% on OSWorld-Verified vs 47.3% for GPT-5.2 and 72.4% for the average human. 1M-token context, 47% token cost reduction in tool-heavy workflows via a new tool-search mechanism, and simultaneous availability across ChatGPT, the API and Codex.

Top stories

• GPT-5.4 launches. First general-purpose OpenAI model with native computer-use; enables agents to operate desktops and carry out complex multi-app workflows. ChatGPT surfaces it as GPT-5.4 Thinking. via OpenAI
• OSWorld-Verified — 75%. Surpasses GPT-5.2’s 47.3% and crosses the average-human threshold (72.4%). via NxCode
• Tool-search mechanism. A new tool-search mode trims token costs by 47% in tool-heavy workflows without measurable accuracy loss. via OpenAI Help

Who shipped

OpenAI with GPT-5.4 across ChatGPT, API, and Codex on day one.

By the numbers

• OSWorld-Verified: 75% (GPT-5.4) vs 47.3% (GPT-5.2) vs 72.4% (average human)
• Context window: 1M tokens
• Tool-search token savings: 47% in tool-heavy workflows

Compiled by AI Feed’s editor from publicly reported announcements on 5 March 2026.