OpenAI Unveils GPT-5.4 With Massive Context Window and Pro Variants
OpenAI releases GPT-5.4 in standard, Thinking, and Pro versions, offering a one-million-token context window and record benchmark scores. The new model significantly reduces hallucinations and introduces a smarter tool-calling system for developers.
OpenAI releases GPT-5.4, calling it its most capable and efficient frontier model for professional work. The new model comes in standard, Thinking, and Pro versions, with an API context window that reaches up to one million tokens. OpenAI emphasizes that GPT-5.4 solves complex problems using significantly fewer tokens than its predecessor.
The new model achieves record scores on several major benchmarks, including OSWorld-Verified, WebArena Verified, and a record 83% on OpenAI's GDPval test for knowledge work tasks. It also leads on Mercor’s APEX-Agents benchmark for law and finance, where it excels at creating long-horizon deliverables like slide decks and financial models while running faster and cheaper than competing models. Additionally, GPT-5.4 reduces individual claim errors by 33% and overall response errors by 18% compared to GPT-5.2.
To improve the developer experience, OpenAI introduces a feature called Tool Search that allows the API to look up tool definitions as needed rather than loading them all into the system prompt. This change results in faster and cheaper requests for systems with many available tools. OpenAI also adds a new safety evaluation for chain-of-thought processes, showing that the Thinking version of GPT-5.4 is less likely to misrepresent its reasoning than previous models.