GPT-5.2 Pro

// Overview

GPT-5.2 Pro is the top reasoning tier of OpenAI's GPT-5.2 family, released on December 11, 2025. Inside ChatGPT, GPT-5.2 ships in three modes — Instant, Thinking, and Pro — where Pro spends the most test-time reasoning compute. OpenAI positions GPT-5.2 Pro as its smartest and most trustworthy model for difficult questions where a higher-quality answer is worth the wait, showing fewer major errors than GPT-5.2 Thinking across complex domains like programming and science.

Technically, GPT-5.2 Pro shares the GPT-5.2 family envelope: a 400,000-token context window, up to 128,000 output tokens, an August 31, 2025 knowledge cutoff, and text plus image (vision) input. It is exposed in the OpenAI API through the Responses API as the model id gpt-5.2-pro. The launch followed an internal 'Code Red' push, arriving roughly three weeks after Google's Gemini 3 Pro.

GPT-5.2 Pro briefly held the verified state-of-the-art on ARC-AGI-2: the ARC Prize team independently confirmed 54.2% for the Pro (High) setting, ahead of GPT-5.2 Thinking's 52.9%. OpenAI has since released GPT-5.4 and GPT-5.5, so GPT-5.2 Pro is now documented as a previous-generation model, but it remains a notable data point for how far extra reasoning compute pushed frontier accuracy in late 2025.

// Benchmarks

1. ARC-AGI-2 (Pro, High — verified)54.2%
2. ARC-AGI-1 (Pro, X-High)90.5%
3. GPQA Diamond (Pro)93.2%
4. AIME 2025 (Pro, no tools)100%
5. Humanity's Last Exam (Pro)36.6%

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

// Pricing

Token prices shown are for the GPT-5.2 family API (gpt-5.2). ARC Prize separately verified a measured cost of $15.72 per task for GPT-5.2 Pro (High) on ARC-AGI-2. OpenAI's current pricing page no longer lists GPT-5.2 after the GPT-5.4 / GPT-5.5 releases; third-party Pro per-token figures disagree, so they are omitted.

Pricing source ↗

// Strengths

• Top verified ARC-AGI-2 result at launch — ARC Prize independently confirmed 54.2% for GPT-5.2 Pro (High), the best on that benchmark at the time
• Highest-effort reasoning tier: spends more test-time compute than GPT-5.2 Thinking for fewer major errors on hard problems
• Strong graduate-level science and competition math — 93.2% GPQA Diamond and 100% on AIME 2025 (vendor-reported)
• Large 400K-token context with 128K max output for long, multi-step professional work
• Vision input — reads charts, screenshots and interface layouts, not just text

// Best for

• Research-grade questions where a slower, higher-quality answer is worth the wait
• Graduate-level science and advanced math reasoning
• Complex, multi-step programming and debugging tasks
• Long-context analysis across large documents (up to ~400K tokens)
• Chart, screenshot and interface understanding via vision input
• Knowledge-work deliverables such as spreadsheets, financial models and presentations

// 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.