OpenAI has introduced GPT-5.4, a new frontier foundation model designed for professional and enterprise use. The company said the release represents its most capable and efficient model to date and is available in multiple configurations, including GPT-5.4 Thinking for advanced reasoning and GPT-5.4 Pro optimized for high-performance workloads.
The API version of GPT-5.4 supports context windows of up to 1 million tokens, enabling significantly larger datasets and longer workflows to be processed in a single interaction. OpenAI also reported improvements in token efficiency, with the model solving tasks using fewer tokens than earlier versions.
Benchmark testing showed strong performance gains across multiple evaluations, including record results on OSWorld-Verified, WebArena Verified, and an 83% score on OpenAI’s GDPval benchmark for knowledge-work tasks. Brendan Foody, CEO of Mercor, said the model demonstrated strong performance on the APEX-Agents benchmark, which evaluates professional capabilities in fields such as law and finance.
OpenAI also reported reduced hallucination rates, with GPT-5.4 showing fewer factual errors compared with earlier models while introducing new tooling and safety evaluations for monitoring reasoning processes.
Featured image: Credit: OpenAI
