GPT-5.4 is here, and it's a significant step forward in what practical AI can do. Released March 5, 2026, this new frontier model from OpenAI combines the best capabilities across reasoning, coding, and professional work into a single, efficient system.

What's New in GPT-5.4?

Native Computer Use: This is the headline feature. GPT-5.4 is the first general-purpose model OpenAI has released with native, state-of-the-art computer-use capabilities. This means agents can operate across applications—use browsers, fill spreadsheets, create presentations—and carry out complex workflows without explicit instruction for each task. It's a step toward fully autonomous AI agents that can actually get work done.

1 Million Token Context: GPT-5.4 supports up to 1M tokens of context. To put that in perspective: you could feed it an entire technical specification, a codebase, and a detailed project brief, and it could reference all of it in reasoning. This is critical for agents planning, executing, and verifying tasks across extended horizons.

Token Efficiency: It's more efficient than GPT-5.2. On GDPval (a benchmark testing agents on 44 professional occupations), GPT-5.4 matches or exceeds professional-level performance in 83% of cases, up from 70.9% for GPT-5.2. On code benchmarks like SWE-Bench Pro, it hits 57.7%, a significant jump. Fewer tokens burned per task means lower costs and faster inference.

The Engineering Details

What makes this tick? GPT-5.4 incorporates the coding capabilities of GPT-5.3-Codex while improving how it works with tools and environments. It includes "tool search"—the ability to identify and use the right tool from an ecosystem without sacrificing reasoning quality. On benchmarks like Toolathlon (testing tool use), it scores 54.6% vs 46.3% for GPT-5.2.

For ChatGPT users, there's a new GPT-5.4 Thinking mode that provides an upfront plan before execution, so you can adjust course mid-response. This reduces back-and-forth on complex queries and maintains context for deep web research.

Why It Matters

The gap between "impressive AI" and "AI that actually does my job" has been closing, but GPT-5.4 feels like a real step across that threshold. Native computer use without custom integration is a game-changer for developers building agentic systems. The efficiency gains and 1M context window mean enterprises can deploy larger, more autonomous systems without proportional cost increases.

This isn't the most advanced model in the abstract sense—but it's arguably the most practical yet. The focus is shifting from "what can AI theoretically do?" to "what can AI reliably do for paying customers?"

Source: OpenAI