Gemini 2.5 Flash (Non-reasoning): benchmark results

Gemini 2.5 Flash evaluated with reasoning disabled. Provider: Google. Released 2025-06-17. Access: API.

Aggregate Rating 1622 ± 7, rank #635 of 2170 rated models, from 582 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AppellateGen - Reversal Prediction Accuracy66.53Ruling-Direction Accuracy (%; affirm vs reverse, balanced 50100
CommunityFact (Closed-Input) - Finance (Japanese)81.6Macro-F1 (%) over True and False claims on the temporally he100
EVGeoQA (Qingdao)26.73Hits@1 (%, times 100) in the No Limit tier (no cap on the di100
GPSBench (Applied) - Relative Position87.2Score (%) on the relative position (compass direction betwee100
GPSBench (Pure GPS) - Relative Position70.8Score (%) on the relative position (compass direction betwee100
TIDE Trajectory - Sudoku - AUV60.4Area Under Variation (0-100)100
TimeSpot - Continent90.51Accuracy (%) of the predicted continent on TimeSpot's 1,455 100
EuroEval Spanish NLU - Sentiment Headlines ES52.95Matthews correlation coefficient (%)97.8
EuroEval Bosnian62.35Average Score (%)97.6
EuroEval Greek NLU - ScaLA EL57.72Matthews correlation coefficient (%)97.5
HELM Arabic - Arabic Mmlu - Natural Science (Middle School)94.63Exact Match (%)97.5
EuroEval Danish Knowledge - Danish Citizen Tests98.35MCC (x100)96.9

Interactive version: theaggregate.ai/model?slug=gemini-2-5-flash-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.