Gemini 1.5 Flash: benchmark results

Google's lightweight, fast Gemini 1.5 tier distilled from 1.5 Pro, with a 1M-token multimodal context (May 2024). Provider: Google. Released 2024-05-15. Access: API.

Aggregate Rating 1548 ± 5, rank #618 of 1656 rated models, from 951 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AudioBench - YouTube ASR Batch 1 WER10.89WER (%) (lower is better)100
AudioBench - YouTube SDS65.97Mean Score (%)100
AudioBench - YouTube SQA78.07Mean Score (%)100
Galileo Tool Tasks - ToolACE Single Function Call99Accuracy (%)100
InstrEditBench (Multi Turn; Two Shot) - DSL BLEU0.05BLEU (0-1) against reference edited text; InstrEditBench DSL100
InstrEditBench (Multi Turn; Two Shot) - DSL ROUGE-L0.12ROUGE-L (0-1) against reference edited text; InstrEditBench 100
InstrEditBench (Multi Turn; Zero Shot) - DSL BLEU0.96BLEU (0-1) against reference edited text; InstrEditBench DSL100
InstrEditBench (Multi Turn; Zero Shot) - DSL ROUGE-L0.98ROUGE-L (0-1) against reference edited text; InstrEditBench 100
PersonaMem v1 - 128K - Acknowledge Latest User Preference59Accuracy (%)100
PersonaMem v1 - 128K - Generalize Reasons to New Scenarios54Accuracy (%)100
RABBITS MedQA Original97.09Accuracy (%)100
VLM-3R Release VSI-Bench Tiny - Abs. Dist.33.6Mean relative accuracy (%)100

Interactive version: theaggregate.ai/model?slug=gemini-1-5-flash · How It Works · Data refreshed daily, snapshot 2026-10-20.