OpenAI Launches GPT-5.4 With Native Computer Use, 1M Token Context, and 33% Fewer Errors

The Biggest GPT Update Since GPT-5

OpenAI just shipped its most significant model update of 2026. GPT-5.4 launched on March 5 in three variants — GPT-5.4 Thinking for ChatGPT subscribers, GPT-5.4 Pro for enterprise and API users, and GPT-5.3 Instant as a lightweight everyday model — consolidating coding, reasoning, and agentic capabilities into a single release.

The headline numbers are impressive. Individual responses are 33 percent less likely to contain errors compared to GPT-5.2, with an overall 18 percent reduction in mistakes across extended conversations. The model tops the leaderboards on Mercor's APEX-Agents benchmark, OSWorld-Verified, and WebArena Verified — all agentic evaluation suites that test real-world task completion rather than synthetic benchmarks.

But the real story isn't the benchmarks. It's what GPT-5.4 can actually do.

Computer Use Goes Mainstream

GPT-5.4 is the first mainline OpenAI model with native computer-use capabilities. That means it can autonomously navigate your desktop, operate applications, issue keyboard and mouse commands, and complete multi-step workflows across different software — all without custom integrations or plugins.

This puts OpenAI in direct competition with Anthropic's computer use feature (launched with Claude 3.5 Sonnet in late 2024) and Google's Project Mariner. The difference is scale: GPT-5.4's computer use ships to ChatGPT's massive user base from day one, not as a research preview.

One Million Tokens of Context

The API version of GPT-5.4 supports context windows up to one million tokens — by far the largest OpenAI has ever offered. For developers, that means entire codebases, full document collections, or months of conversation history can be processed in a single request.

GPT-5.4 is also the first OpenAI model trained with what the company calls "compaction" — a technique for preserving key context during extended agent operations. When the model hits context limits during long-running tasks, it intelligently summarizes and retains the most relevant information rather than simply truncating.

Six Key Improvements

OpenAI highlighted six areas of advancement: stronger coding and document understanding, enhanced multimodal image perception, better long-running agent workflows, improved token efficiency for tool-heavy workloads, advanced agentic web search with multi-source synthesis, and enhanced business workflows for document and spreadsheet-heavy processes.

The business focus is deliberate. ChatGPT for Excel — a new add-in powered by GPT-5.4 — launched alongside the model, bringing AI directly into workbooks to build financial models, run scenarios, and generate outputs from cell data. It's a clear signal that OpenAI sees enterprise productivity as the next major revenue driver.

Availability

GPT-5.4 Thinking is rolling out now for ChatGPT Plus, Teams, and Pro subscribers. GPT-5.4 Pro is available through the API and for ChatGPT Enterprise and Edu users. The model will fully replace GPT-5.2 Thinking within three months.

Sources: TechCrunch (March 5, 2026), 9to5Mac (March 5, 2026), Gizmodo (March 6, 2026)