OpenAI (AI company)

OpenAI

OpenAI

to release / to launch

fābù

发布

GPT-5.4 (AI language model)

GPT-5.4

GPT-5.4

，

reasoning ability

tuīlǐ

推理

and / with

yǔ

与

programming / coding

biānchéng

编程

capability / ability

nénglì

能力

significantly / by a large margin

dàfú

大幅

to improve / to upgrade

tíshēng

提升

。

New Words

News in English

OpenAI Releases GPT-5.4 With Major Gains in Reasoning and Coding

OpenAI launched GPT-5.4 on March 5, 2026, its most capable frontier language model yet, with major advances in reasoning, coding, and agentic tasks. The new model reduces factual error rates by 33 percent, and lightweight versions GPT-5.4 mini and nano were launched later that month.

On March 5, 2026, American artificial intelligence company OpenAI officially launched the next-generation large-scale language model GPT-5.4.

This release includes two versions: GPT-5.4 Thinking for reasoning-focused tasks and GPT-5.4 Pro for professional applications. Both models are accessible through ChatGPT and API platforms.

OpenAI stated that GPT-5.4 demonstrates significant improvements in reasoning capabilities, programming tasks, and complex workflows.

Compared to the previous version, GPT-5.4 reduced the factual error rate by 33 percent, with overall response quality becoming even more stable.

The model's context window can reach up to one million tokens, making it the largest context capacity OpenAI has released to date.

The model also supports more complex AI agent operations and can collaboratively complete professional tasks across multiple software environments.

On March 17, OpenAI further launched lightweight versions: GPT-5.4 mini and GPT-5.4 nano, primarily aimed at high-throughput commercial applications.

Analysts believe this series of releases consolidated OpenAI's leading position in global AI competition.

More in Science & Technology