Reka Flash 3 — benchmark results
Reka's open-weights 21B reasoning model trained from scratch (Apache 2.0, 32K context), aimed at low-latency and on-device use (March 2025). Provider: Reka. Released 2025-03-10. Access: Open.
Unified ELO 1529 ± 14, rank #688 of 1776 rated models, from 152 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval German NLU - Sb10K | 59.6 | Sentiment classification Score (%) | 93.1 |
| EuroEval French NLU - Eltec | 71.29 | Named entity recognition Score (%) | 92.8 |
| EuroEval Danish NLU - Angry Tweets | 57.97 | Sentiment classification Score (%) | 92.2 |
| EuroEval English NLU - SST-5 | 69.19 | Sentiment classification Score (%) | 91.9 |
| EuroEval Italian NLU - Sentipolc16 | 65.9 | Sentiment classification Score (%) | 90.5 |
| EuroEval Spanish NLU - CoNLL ES | 75.76 | Named entity recognition Score (%) | 89.8 |
| EuroEval English NLU - CoNLL EN | 81.67 | Named entity recognition Score (%) | 88 |
| EuroEval German NLU - GermEval | 68.07 | Named entity recognition Score (%) | 82.9 |
| EuroEval Portuguese NLU - MultiWikiQA PT | 74.48 | Reading comprehension Score (%) | 82 |
| EuroEval Dutch | 67 | Average Score (%) | 81.2 |
| EuroEval Portuguese NLU - HAREM | 51.85 | Named entity recognition Score (%) | 81.2 |
| EuroEval Dutch NLU | 71.73 | NLU Average Score (%) | 80.8 |
Interactive version: theaggregate.ai/model?slug=reka-flash-3 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.