Gemini 3.5 Flash: benchmark results

Google's speed-oriented Gemini 3.5 Flash model, balancing capability with lower latency. Provider: Google. Released 2026-05-19. Access: API.

Unified ELO 1726 ± 1, rank #28 of 1392 rated models, from 268 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
APEX-Agents66.1Mean Score (ReAct) (self-reported)100
Arabic Broad Leaderboard9.25Average Score (0-10)100
CAIS Vision Capabilities Index65.9Vision Capabilities Index (self-reported)100
LiveMathematicianBench49.5Accuracy (%)100
PerceptionBench - Hallucination50.6Perception Hallucination (%)100
SciCode-Verified97.2Subproblem Resolve Rate (%)100
Nejumi 4 - ALT - Ethics & Morality99Score (%)99.6
Nejumi 4 - GLP - General Knowledge85.27Score (%)99.2
Roboflow Vision Evals - Visual Understanding79.1Pass Rate (%)98.7
AI for Education Visual Reasoning86Accuracy (%)98.5
AI for Education Visual Reasoning - match (figure)85.2Accuracy (%)98.5
AI for Education Visual Reasoning - reasoning by analogy88.8Accuracy (%)98.5

Interactive version: theaggregate.ai/model?slug=gemini-3-5-flash · How It Works · Data refreshed daily, snapshot 2026-09-05.