GPT-5.4 Arrives: OpenAI's Unified AI Model Can Now Control Your Computer and Outthink Your Coworkers

March 06, 2026
GPT-5.4
4 min

Breaking: GPT-5.4 Goes Live Across ChatGPT, API, and Codex

On Thursday, March 5, 2026 (EST), OpenAI officially launched GPT-5.4, its most capable and token-efficient frontier model to date. The release rolled out simultaneously across ChatGPT, the developer API, and the Codex platform. OpenAI CEO Sam Altman teased the launch on X (formerly Twitter), writing: "I think people will like this."

The new model is available immediately to ChatGPT Plus, Team, and Pro subscribers. Enterprise and Education plan users can enable early access through admin settings. A more powerful variant, GPT-5.4 Pro, is also available for users who require maximum performance on highly complex tasks.

What's New: A Unified Powerhouse Model

GPT-5.4 represents a major consolidation in OpenAI's model lineup. It merges the industry-leading coding capabilities of GPT-5.3-Codex with enhanced reasoning, agentic workflows, and professional productivity tools — all in a single model.

Key upgrades include:

  • Native Computer-Use Capabilities: For the first time in a general-purpose model, GPT-5.4 in Codex and the API can autonomously operate computers, navigate browsers and desktop applications, and carry out complex multi-step workflows.
  • 1 Million Token Context Window: GPT-5.4 supports up to 1 million tokens of context, enabling agents to plan, execute, and verify tasks across extended sessions.
  • Tool Search System: A newly introduced Tool Search feature allows the model to look up tool definitions only when needed, reducing token usage and improving response speed in large tool ecosystems.
  • Upfront Reasoning Plans: In ChatGPT, the GPT-5.4 Thinking version can present an initial plan of its reasoning before generating the full response, letting users adjust course mid-process.
  • Token Efficiency: GPT-5.4 is OpenAI's most token-efficient reasoning model to date, using significantly fewer tokens than GPT-5.2 to solve problems — helping offset the slightly higher per-token price.

Benchmark Performance: Record-Breaking Results

GPT-5.4 shattered performance benchmarks across multiple professional evaluation frameworks:

  • GDPval: Scored 83%, outperforming office workers across 44 occupations on real-world tasks.
  • APEX-Agents (Mercor): Achieved top ranking on this benchmark designed to test AI performance in law and finance.
  • OSWorld-Verified & WebArena Verified: Set new records on computer-use benchmarks that measure how effectively AI systems interact with software environments.
  • Spreadsheet Modeling: Scored 87.3% on an internal benchmark simulating investment banking analyst tasks, versus 68.4% for GPT-5.2.
  • Presentation Generation: Human raters preferred GPT-5.4's presentations 68% of the time over GPT-5.2's outputs.

Reduced Hallucinations and Improved Accuracy

OpenAI placed a strong emphasis on reliability in this release. According to the company, GPT-5.4 is:

  • 33% less likely to produce errors in individual factual claims compared to GPT-5.2.
  • 18% less likely to have overall responses containing factual mistakes.

A new safety evaluation focused on chain-of-thought (CoT) reasoning found that deception is less likely in the GPT-5.4 Thinking version, with OpenAI stating that "the model lacks the ability to hide its reasoning and that CoT monitoring remains an effective safety tool."

Enterprise and Competitive Implications

The launch signals OpenAI's intensifying push into the enterprise market — a space where Anthropic has historically held a strong position. GPT-5.4's out-of-the-box agentic capabilities, combined with its professional document, spreadsheet, and presentation skills, put it in direct competition with Anthropic's Claude for enterprise workflows.

Alongside GPT-5.4, OpenAI also debuted a ChatGPT for Excel add-in, bringing AI directly into Microsoft's ubiquitous spreadsheet software. New app integrations and skills were also announced for use within ChatGPT.

Market analysts are watching closely. Earlier in 2026, the release of Anthropic's Cowork plug-ins triggered a broad selloff in SaaS stocks. A similar reaction may follow as GPT-5.4's agentic capabilities raise fresh questions about the future of enterprise software.

Model Availability and Legacy Transition

  • GPT-5.4 Thinking is now live in ChatGPT for Plus, Team, and Pro users, replacing GPT-5.2 Thinking.
  • GPT-5.2 Thinking will remain available in the Legacy Models section for three months, before being retired on June 5, 2026 (EST).
  • API pricing on OpenRouter is listed at $2.50 per 1M input tokens and $20.00 per 1M output tokens, with a 1M context window and 128K max output.
  • Prompts exceeding 272K input tokens are subject to 2x input and 1.5x output pricing for the full session.

Bottom Line

GPT-5.4 is OpenAI's most comprehensive model release in recent memory — combining frontier reasoning, coding, computer-use autonomy, and professional productivity into a single, more efficient package. With record benchmark scores, significant hallucination reductions, and native enterprise integrations, it sets a new standard and intensifies the race among the world's leading AI labs.