GPT-5.4 is OpenAI's Push Toward Agentic Frontier Models for Professional Work

Ronni Holmvig Strøm · 2026-03-06

OpenAI released GPT-5.4 on March 5, 2026, billing it as the company's "most capable and efficient frontier model for professional work." Available immediately in ChatGPT (as GPT-5.4 Thinking), the API (as gpt-5.4 and gpt-5.4-pro), and Codex, the update arrives just two days after GPT-5.3 Instant and continues OpenAI's accelerated cadence in 2026.

OpenAI released GPT-5.4 on March 5, 2026, billing it as the company's "most capable and efficient frontier model for professional work."

Available immediately in ChatGPT (as GPT-5.4 Thinking), the API (as gpt-5.4 and gpt-5.4-pro), and Codex, the update arrives just two days after GPT-5.3 Instant and continues OpenAI's accelerated cadence in 2026.

This isn't a full generational leap like the original GPT-5 (August 2025). Instead, GPT-5.4 unifies and refines recent advances in reasoning, coding, tool use, and agentic workflows into a single, more token-efficient flagship—positioned explicitly as a direct competitor to enterprise-focused rivals like Anthropic's Claude series.

Key Capabilities and Variants

OpenAI ships GPT-5.4 in two main flavors:

GPT-5.4 Thinking (default in ChatGPT for Plus, Team, Pro users; Enterprise/Edu via early access): Optimized for multi-step reasoning, long-context tasks, tool-heavy workflows, and knowledge work. It replaces GPT-5.2 Thinking (legacy access until June 5, 2026).

GPT-5.4 Pro (API as gpt-5.4-pro; ChatGPT Pro/Enterprise): Higher-compute variant for maximum performance on the most complex problems.

Standout features include:

Native computer-use capabilities. GPT-5.4 is OpenAI's first general-purpose model with built-in ability to operate computers: interpreting screenshots, issuing keyboard/mouse commands, navigating software, and executing multi-step actions across applications.

This builds on earlier agent prototypes and positions the model as a foundation for autonomous software agents.

1 million token context window (API/Codex). The largest from OpenAI to date, enabling agents to plan, execute, and verify over massive documents, codebases, or datasets. Pricing doubles beyond 272k tokens (input 2×, output 1.5×).

Token efficiency gains. Up to 47% fewer tokens on some tasks vs. GPT-5.2, reducing cost and latency for iterative agentic loops.