Gemini 1.5 Flash-8B: benchmark results

Google's smallest, cheapest Gemini 1.5 tier (GA October 2024), an 8B Flash variant with a 1M context for high-volume tasks. Provider: Google. Released 2024-10-03. Access: API.

Unified ELO 1456 ± 1, rank #917 of 1392 rated models, from 96 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
MERA - ruCodeEval61.28pass@1 (%)90.4
AV-Odyssey - Audio Time Estimation31Score (%)88.2
MERA - ruHumanEval58.41pass@1 (%)87.5
AV-Odyssey - Melody28.9Score (%)82.4
AV-Odyssey - Music Genre Classification33Score (%)82.4
AV-Odyssey - Animal Recognition28Score (%)76.5
AV-Odyssey - Action Tracing34.9Score (%)73.5
MERA - BPS98.2Accuracy (%)73.1
MERA - ruDetox29.4Joint Score (%)72.1
RewardBench76.01Score (%)70.7
AV-Odyssey - Action Recognition32.7Score (%)70.6
AV-Odyssey - Music Sentiment Analysis25.9Score (%)70.6

Interactive version: theaggregate.ai/model?slug=gemini-1-5-flash-8b · How It Works · Data refreshed daily, snapshot 2026-09-05.