GPT-5.4 Pro

// Overview

GPT-5.4 Pro is the top reasoning tier of OpenAI's GPT-5.4 family, launched on March 5, 2026 alongside GPT-5.4 Thinking. Where the standard GPT-5.4 balances speed and cost, GPT-5.4 Pro is tuned to use additional inference-time compute to "think harder" and return more consistent answers on the hardest problems. OpenAI positions the GPT-5.4 line as its most capable frontier model for professional work, with strengths in long-horizon deliverables such as slide decks, financial models, and legal analysis.

GPT-5.4 Pro accepts text and image input and supports a 1.05M-token context window with up to 128K output tokens, making it suited to long-context workflows that span large codebases, document sets, and multi-step agentic tasks. It exposes adjustable reasoning effort (medium, high, xhigh) and is available only through the Responses API, with native tool use including web search, file search, image generation, computer use, and MCP. Because a single request can take minutes of reasoning, OpenAI recommends background mode to avoid timeouts.

In ChatGPT, GPT-5.4 Pro is reserved for Pro and Enterprise plans. Via the API it is priced as a premium reasoning tier at $30 per million input tokens and $180 per million output tokens. On evaluations, the GPT-5.4 line set state-of-the-art results in computer use and professional knowledge work, and GPT-5.4 Pro pushes science and agentic-search scores higher still (94.4% GPQA Diamond, 89.3% BrowseComp).

// Benchmarks

1. GPQA Diamond94.4%
2. BrowseComp89.3%
3. ARC-AGI-2 (Verified)83.3%
4. FrontierMath Tier 438%
5. GDPval (GPT-5.4)83%
6. OSWorld-Verified (GPT-5.4)75%
7. SWE-Bench Pro, Public (GPT-5.4)57.7%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

ChatGPT Pro/Enterprise tier; API via the Responses API only. Prompts over 272K input tokens incur a 2x input / 1.5x output multiplier.

Pricing source ↗

// Strengths

• Maximum-effort reasoning for high-stakes, complex tasks where consistency matters more than latency
• Very large 1.05M-token context window for long-context and whole-repo workflows
• State-of-the-art expert science reasoning (94.4% on GPQA Diamond)
• Strong agentic web search and deep research (89.3% on BrowseComp)
• Native tool use: web search, file search, image generation, computer use, and MCP
• Lower error rate than GPT-5.2 (OpenAI reports 33% fewer errors in individual claims)

// Best for

• Long-horizon professional deliverables: financial models, slide decks, and legal analysis
• Expert-level science, math, and research questions that need careful reasoning
• Deep research and multi-round agentic web search over many sources
• Agentic coding and multi-step problem solving across large codebases
• Long-context document and codebase analysis using the 1.05M-token window
• Computer-use automation that drives a desktop or browser to complete tasks

// How to access

// GPT Pro — every version

The full lineage of the GPT Pro line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.