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.

Unified ELO 1492 ± 7, rank #829 of 1776 rated models, from 291 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EuroEval French NLU - ScaLA FR59.35Linguistic acceptability Score (%)92.6
EuroEval Italian NLU - ScaLA IT47.66Linguistic acceptability Score (%)91.2
EuroEval English NLU - ScaLA EN60.57Linguistic acceptability Score (%)90.7
EuroEval Portuguese NLU - ScaLA PT42.68Linguistic acceptability Score (%)90.7
EuroEval Spanish NLU - ScaLA ES40.12Linguistic acceptability Score (%)89.8
EuroEval English NLU71.36NLU Average Score (%)86.4
EuroEval French NLU68.43NLU Average Score (%)85.3
EuroEval English Knowledge90.65Knowledge Average Score (%)83.9
EuroEval Danish NLU - Angry Tweets55.5Sentiment classification Score (%)82.5
EuroEval Romanian Knowledge68.52Knowledge Average Score (%)81.5
EuroEval Bosnian52.97Average Score (%)81
EuroEval English71.69Average Score (%)80.4

Interactive version: theaggregate.ai/model?slug=trinity-mini · How the rankings work · Data refreshed daily, snapshot 2026-07-22.