Gemini 2.5 Pro (High): benchmark results

Provider: Google. Released 2025-03-25. Access: API.

Aggregate Rating 1669 ± 17, rank #471 of 2257 rated models, from 78 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EpiCurveBench52.3EpiCurveSimilarity, ECS (%): edit distance with real penalty100
EpiCurveBench - High Reasoning with Code Execution48EpiCurveSimilarity, ECS (%): edit distance with real penalty100
EpiCurveBench - Manually Annotated Set56.9EpiCurveSimilarity, ECS (%): edit distance with real penalty100
Misguided Attention - feeding_the_goat1Mean native expected-behavior score (0–1)94.6
VTC-Bench (Interface Tools) - Perceptual Restoration40Pass rate (%) on perceptual restoration (50 questions under 94.4
Misguided Attention - lonely_rabbit_impossible1Mean native expected-behavior score (0–1)92
VTC-Bench (Interface Tools) - Measurement44.76Pass rate (%) on measurement (105 questions on size, positio91.7
Misguided Attention - bridge_torch_easy1Mean native expected-behavior score (0–1)91.1
VTC-Bench (Code Tools) - Attention Focusing40Pass rate (%) on attention focusing (45 questions on rotated88.9
VTC-Bench (Code Tools) - Robust OCR46Pass rate (%) on robust OCR (50 questions on degraded text),88.9
VTC-Bench (Interface Tools) - Attention Focusing44.44Pass rate (%) on attention focusing (45 questions on rotated88.9
VTC-Bench (Interface Tools) - Robust OCR60Pass rate (%) on robust OCR (50 questions on degraded text),88.9

Interactive version: theaggregate.ai/model?slug=gemini-2-5-pro-high · How It Works · Data refreshed daily, snapshot 2026-11-17.