OpenAI launches GPT-5.4 with record computer control and 1 million token context window

Quick Facts

• GPT-5.4 sets industry record with 75% score on OSWorld computer control benchmark, beating human testers
• Model offers 1 million token context window and reduces token usage by 47% compared to GPT-5.2
• Available in three variants with API pricing at $2.50/M input tokens and $15/M output tokens

OpenAI released GPT-5.4 on March 5, 2026, marking a significant advancement in computer control capabilities. The model achieved a 75% score on the OSWorld-Verified benchmark, surpassing both GPT-5.2 and the 72.4% typically achieved by human testers.

The launch includes three model variants: standard GPT-5.4, GPT-5.4 Thinking for reasoning tasks, and GPT-5.4 Pro for high-performance applications. The API version supports context windows up to 1 million tokens, the largest offered by OpenAI.

GPT-5.4 demonstrated substantial efficiency improvements, using 47% fewer tokens than previous versions while maintaining the same accuracy on Scale’s MCP Atlas benchmark. The model also showed 33% fewer errors in individual claims and 18% fewer overall response errors compared to GPT-5.2.

“We’re seeing GPT-5.4 perform exceptionally well at logical reasoning and executing intricate, multi-step, tool-dependent workflows,” said Mario Rodriguez, Chief Product Officer at GitHub.

The model achieved an 83% win rate on knowledge work tasks through the GDPval benchmark, jumping from GPT-5.2’s 70.9% score. On WebArena-Verified, GPT-5.4 reached 67.3% success, and 92.8% on Online-Mind2Web using screenshot-based observations.

OpenAI CEO Sam Altman highlighted the business potential: “I think a company like OpenAI will be able to build a $100 billion business line, just selling products to other businesses. The efficiency gains, the value delivered is so worth it.”

Enterprise adoption has surged since launch, with API usage processing more than twice as much coding and agent-building work. Reasoning use cases jumped eightfold, particularly for planning and multi-step reasoning tasks.

The launch intensifies competition with Anthropic’s Claude for Financial Services and Google’s Gemini 3.1 Pro. ChatGPT maintains 68% market share, down from 87.2% a year ago, while Google Gemini captured 18.2% compared to 5.4% in January 2025.

GPT-5.4 introduces native computer-use capabilities, enabling agents to operate computers and execute multi-step workflows across applications. A new search engine automatically finds required tools, reducing prompt sizes and inference costs.

Read more: OpenAI launches GPT-5.4 with computer vision, tool use enhancements