What’s new in OpenAI’s GPT‑5.4?
The capabilities OpenAI added and why they matter
OpenAI’s latest frontier model brings a string of new features aimed at professional workflows. The company released both GPT‑5.4 Thinking and GPT‑5.4 Pro variants, and emphasized improvements across reasoning, tool use, and long‑form context. One headline capability is native computer control: the model can perform operations on a desktop environment and invoke tools more directly than prior releases. That is paired with an expanded context window that reaches up to one million tokens for the API, letting the model work with far larger documents or multimodal project states without losing earlier context.
Other notable product changes
- Enhanced tool‑calling and integration, including new spreadsheet and financial plugins for Microsoft Excel and Google Sheets.
- Presentation and image‑generation improvements: outputs now produce stronger aesthetics and more varied slide layouts for professional decks.
- Accuracy and safety claims: OpenAI reports that individual factual claims are materially less likely to be false and that full responses contain fewer errors compared with the previous frontier model.
- Pricing tiers: the company published per‑token pricing for both standard and Pro offerings, reflecting different cost/latency trade‑offs for heavy professional use.
Why this matters
These updates move the product beyond conversational Q&A toward acting as a hands‑on assistant for knowledge work—drafting reports, manipulating spreadsheets, and orchestrating toolchains. That raises productivity possibilities for professionals but also concentrates new operational risks: native computer use and broader tool access make governance, auditability, and secure deployments more urgent for enterprises. The changes also intensify competition: the model’s agentic features and expanded context put pressure on rivals to match both capability and enterprise controls.