Trinity-Mini: benchmark results
Arcee AI's US-trained 26B sparse-MoE (3B active) reasoning and tool-use model, first release of its from-scratch AFMoE Trinity family (December 2025). Provider: Arcee AI. Released 2025-12-02. Access: Open.
Aggregate Rating 1443 ± 13, rank #1250 of 2133 rated models, from 302 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LatamBoard | 87.13 | LatamBoard Category Average (percentage) | 94.9 |
| EuroEval French NLU - ScaLA FR | 59.35 | Linguistic acceptability Score (%) | 92.6 |
| EuroEval Italian NLU - ScaLA IT | 47.66 | Linguistic acceptability Score (%) | 91.2 |
| EuroEval English NLU - ScaLA EN | 60.57 | Linguistic acceptability Score (%) | 90.7 |
| EuroEval Portuguese NLU - ScaLA PT | 42.68 | Linguistic acceptability Score (%) | 90.7 |
| EuroEval Spanish NLU - ScaLA ES | 40.12 | Linguistic acceptability Score (%) | 89.8 |
| EuroEval English NLU | 71.36 | NLU Average Score (%) | 86.4 |
| EuroEval French NLU | 68.43 | NLU Average Score (%) | 85.3 |
| EuroEval English Knowledge | 90.65 | Knowledge Average Score (%) | 83.9 |
| EuroEval Danish NLU - Angry Tweets | 55.5 | Sentiment classification Score (%) | 82.5 |
| EuroEval Romanian Knowledge | 68.52 | Knowledge Average Score (%) | 81.5 |
| EuroEval Bosnian | 52.97 | Average Score (%) | 81 |
Interactive version: theaggregate.ai/model?slug=trinity-mini · How It Works · Data refreshed daily, snapshot 2026-10-13.