AFM-4.5B: benchmark results
Arcee AI's AFM-4.5B instruct model, pretrained on 8 trillion tokens (including 1.5 trillion focused on math and code) then refined with supervised fine-tuning and reinforcement learning, for cloud-to-edge deployment. Provider: Arcee AI. Released 2025-09-01. Access: Open.
Aggregate Rating 1355 ± 16, rank #1372 of 1656 rated models, from 309 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LatamBoard - Spanish Escola | 71.32 | Score (%) | 93.9 |
| Diplomacy: Betrayal Tendency | 0 | Betrayal rate (%), 60 critical-state games resumed at Fall 1 | 87.5 |
| EuroEval English NLU - ScaLA EN | 52.46 | Matthews correlation coefficient (%) | 79 |
| Kurakurai French LLM - GPQA Diamond | 29.6 | Accuracy (%) | 78 |
| LatamBoard - FaQuAD NLI | 75.49 | Score (%) | 75.8 |
| LatamBoard - Spanish WNLI | 74.65 | Score (%) | 74.2 |
| EuroEval English Knowledge | 81.61 | Knowledge Average Score (%) | 66.5 |
| EuroEval French NLU - ScaLA FR | 36.8 | Matthews correlation coefficient (%) | 59.2 |
| EuroEval Spanish NLU - ScaLA ES | 18.77 | Matthews correlation coefficient (%) | 55.1 |
| Kurakurai French LLM - AIME 2025 | 0.67 | Accuracy (%) | 54.9 |
| Kurakurai French LLM - Global MMLU | 49.8 | Accuracy (%) | 53.7 |
| EuroEval English NLU - SST-5 | 65.03 | Matthews correlation coefficient (%) | 53 |
Interactive version: theaggregate.ai/model?slug=afm-4-5b · How It Works · Data refreshed daily, snapshot 2026-10-20.