GPT-5.2 Pro — benchmark results

OpenAI's higher-compute GPT-5.2 variant for difficult reasoning and general tasks. Provider: OpenAI. Released 2025-12-11. Access: API.

Unified ELO 1926 ± 27, rank #51 of 1776 rated models, from 59 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
LLM Stats (HMMT 2025)100Score (%)100
SuperGPQA67.13Accuracy (%)98.5
ZeroEval GPQA Diamond93.2GPQA Diamond Score96.9
Pencil Puzzle Bench - Sudoku6.7Direct-ask Success Rate (%)96
SGI-Bench Idea Generation55.03Idea Generation Score94.4
FrontierMath - Tier 431.3Accuracy (%, 48 problems)93
IUMB89.6Score (%)92.7
Pencil Puzzle Bench - Norinori53.3Direct-ask Success Rate (%)91
Pencil Puzzle Bench - Nurimisaki20Direct-ask Success Rate (%)91
DEEPSYNTH8.7F1 (self-reported)90
Pencil Puzzle Bench - Hitori26.7Direct-ask Success Rate (%)89
Pencil Puzzle Bench - Nurimaze6.7Direct-ask Success Rate (%)89

Interactive version: theaggregate.ai/model?slug=gpt-5-2-pro · How the rankings work · Data refreshed daily, snapshot 2026-07-22.