OpenAI releases GPT-5.4, including Pro and Thinking versions.

CoinNetwork · 2026-03-05T21:50:57+00:00

OpenAI has released the new flagship model GPT-5.4, optimized for reasoning and high performance, offering a context window of 1 million tokens and improving token efficiency. It performs excellently across multiple benchmark tests, reducing the rate of personal factual errors, and introduces new tool search systems to enhance the safety of logical reasoning chains.

CoinNetwork

2026-03-05 21:50:57

Abstract generation in progress

OpenAI has released a new flagship model for professional work, GPT-5.4, which includes optimized versions for reasoning and high performance (Pro). Its API offers a 1 million token context window and improved token efficiency. GPT-5.4 set new records on benchmarks such as OSWorld-Verified and WebArena Verified, scored 83% on OpenAI’s GDPval tests, and ranked highly on Mercor’s APEX-Agents benchmarks in the legal and financial fields. Compared to GPT-5.2, the model reduces factual errors in personal statements by 33%. Additionally, GPT-5.4 introduces a new tool search system for API tool calls and enhances safety assessments for logical reasoning chains.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.