Gemini 3 Flash (Preview) (Non-reasoning): benchmark results
Gemini 3 Flash (Preview) evaluated with reasoning disabled. Provider: Google. Released 2025-12-17. Access: API.
Aggregate Rating 1688 ± 11, rank #394 of 2170 rated models, from 329 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentProcessBench - HotPotQA Step Accuracy | 76.7 | StepAcc (%): share of assistant steps, pooled over trajector | 100 |
| Doc-PP - Implicit Query Faithfulness (With Policy) | 69 | Faithfulness (%; content-checklist criteria met, implicit qu | 100 |
| EuroEval Albanian NLU - MMS SQ | 32.13 | Matthews correlation coefficient (%) | 100 |
| EuroEval Bulgarian Knowledge | 80.16 | Knowledge Average Score (%) | 100 |
| EuroEval Czech Summarization - Czech News | 30.2 | chrF3++ (%) | 100 |
| EuroEval Estonian NLU - Grammar ET | 66.32 | Matthews correlation coefficient (%) | 100 |
| EuroEval Finnish Summarization - Xlsum FI | 32.86 | chrF3++ (%) | 100 |
| nsanku-MMLU - Akuapem Twi | 66.5 | Accuracy (%) | 100 |
| nsanku-MMLU - Asante Twi | 68 | Accuracy (%) | 100 |
| nsanku-MMLU - Dagaare | 47 | Accuracy (%) | 100 |
| nsanku-MMLU - Dagbani | 47 | Accuracy (%) | 100 |
| nsanku-MMLU - Dangme | 47 | Accuracy (%) | 100 |
Interactive version: theaggregate.ai/model?slug=gemini-3-flash-preview-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.