What does GPT‑5.4 change?
New model pushes toward professional agents
OpenAI released GPT‑5.4 as a step aimed at heavier knowledge‑work and agentic tasks. The company positioned this release as its most capable frontier model for professional uses, introducing features designed to let the model do more complex, longer‑running work and to integrate more directly with users’ tools.
Key technical and product changes:
- Native computer control: The model can interact with files, run certain tools, and automate workflows on behalf of users, opening a path toward more autonomous agent behaviors.
- Much larger context: APIs and product notes indicate support for context windows up to around one million tokens, letting it handle longer documents and extended conversations.
- Improved tool calling: The tool‑calling system was reworked to make integrations more reliable when the model needs to reach out to external services or plugins.
- Spreadsheet and workflow integrations: New capabilities bring direct ChatGPT functionality into Excel and Google Sheets, plus features aimed at finance and professional workflows.
What users and businesses should watch
- Productivity gains: Firms that rely on document synthesis, reporting, or spreadsheet work could see immediate boosts if the model reliably reduces manual tasks.
- Cost and governance: The new capabilities come with higher potential compute costs and the need for stricter controls around data access, tool permissions, and auditability.
- Safety and competition: OpenAI reports measurable reductions in factual errors versus previous versions, but the expanded autonomy raises fresh safety questions. The release also intensifies competition with rivals positioning their own agent‑capable models.
Availability and pricing notes
GPT‑5.4 is available in multiple tiers (standard and Pro/Thinking variants) with tiered pricing that reflects the model’s professional focus. Exact pricing varies by token usage and service tier; organizations should weigh performance gains against costs and governance overhead before broad deployment.