Atlas

Models

← All models

GPT-5.2 Pro

OpenAI · 2025-12-11

GPT-5.2 Pro is the highest-capability variant of OpenAI's GPT-5.2 family, released on December 11, 2025 for difficult questions where higher-quality answers justify longer processing time. It supports long-context understanding, agentic tool use, and vision, and is available in ChatGPT and through the API.

Benchmark scores

BenchmarkScore
ARC-AGI-1 Semi-Private90.5
ARC-AGI-238.5
BrowseComp77.9
BuseyBench SVG5.5
Capability157.1
Chess Puzzles (task version 2.0)60.0
Epoch Capabilities Index (ECI)156
Eval Connections100.0
Eval Connections — Guess Accuracy100.0
Extended NYT Connections 940 (Quadratic v1)79.3
Extended NYT Connections 940-Puzzle Set85.7
Extended NYT Connections Newest 10080.8
Extended NYT Connections Newest 100 (Quadratic v1)71.8
Fiction.LiveBench - 0 tokens (relevant passages only)100.0
Fiction.LiveBench - 120K tokens75.0
Fiction.LiveBench - 16K tokens83.3
Fiction.LiveBench - 192K tokens78.1
Fiction.LiveBench - 1K tokens100.0
Fiction.LiveBench - 2K tokens91.7
Fiction.LiveBench - 32K tokens80.6
Fiction.LiveBench - 400 tokens100.0
Fiction.LiveBench - 4K tokens80.6
Fiction.LiveBench - 60K tokens80.6
Fiction.LiveBench - 8K tokens88.9
FrontierMath Tier 4 (v1) - Private31.3
FrontierMath Tier 4 (v2) - Private46.0
FrontierMath Tiers 1-3 (v2) - Private74.0
GDPval74.1
GeneBench-Pro8.5
GeneBench-Pro Artificial Analysis Subset2.4
Loading Atlas data…