Article: OpenAI introduces GPT-5.4 with more knowledge-work capability

Meridiano

Add Article

Article Details

Impact Score

6 / 10

AI Summary (Processed Content)

OpenAI has released GPT-5.4, a model update focused on agentic tasks and computer-use capabilities, allowing it to control a computer via keyboard and mouse inputs. The new "Thinking" variant shows more of its reasoning process and can adjust mid-task, improving performance on long-horizon work and research.

The update also brings significant efficiency gains and an expanded API context window of 1 million tokens, matching competitors. This release aims to retain users amid recent shifts to rival models from Anthropic and Google.

Main topics: OpenAI's GPT-5.4 release, its focus on computer-use and agentic tasks, reasoning and efficiency improvements, and competitive positioning.

Original URL

https://arstechnica.com/ai/2026/03/openai-introduces-gpt-5-4-with-more-knowledge-work-capability/

Source Feed

Ars Technica

Published Date

2026-03-05 20:55

Fetched Date

2026-03-05 18:30

Processed Date

2026-03-05 18:31

Embedding Status

Present

Cluster ID

Not Clustered

Raw Extracted Content

In keeping with its recently accelerated release cadence, OpenAI has shipped GPT-5.4 (including GPT-5.4 Thinking and GPT-5.4 Pro).
This update comes at a critical time, as recent events have led some vocal users to abandon ship for competing products and models from Anthropic and Google.
GPT-5.4 is another model update focused on usefulness for agentic tasks, particularly knowledge work. OpenAI says this is its first model explicitly aimed at computer-use tasks; like competing models, it can issue keyboard or mouse inputs based on periodic desktop or application screenshots.
When prompted within ChatGPT, the GPT-5.4 Thinking model will present more of its reasoning up front and can accept prompts to change course in the middle of its reasoning process, OpenAI says. It’s said to improve context maintenance for long, so-called reasoning stretches. That makes it more useful for long-horizon tasks and web research.
Also helpful for long-horizon tasks is improved token efficiency—as is often the case with these new model releases, efficiency improvements allow it to do more before users hit limits. On the API side, the context window has been increased to 1 million tokens, competing with offerings from Google and Anthropic.