SmolLM2-360M-Instruct: benchmark results
Hugging Face's SmolLM2-360M-Instruct, fine-tuned from SmolLM2-360M (trained on 4 trillion tokens) via supervised fine-tuning and DPO on UltraFeedback, improving on SmolLM1 in instruction following and reasoning. Provider: HuggingFace. Released 2024-10-31. Access: Open.
Aggregate Rating 1161 ± 16, rank #1595 of 1635 rated models, from 324 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Enkrypt AI - Bias Risk | 66.41 | Risk Score | 81.6 |
| SLM Trustworthiness - Fairness | 47.45 | TrustLLM fairness accuracy (%; sub-task mean) | 70 |
| SLM Trustworthiness - Privacy | 48.59 | TrustLLM privacy accuracy (%; sub-task mean) | 60 |
| MEDIC Benchmark | 62.47 | MEDIC Public Table Average | 50.5 |
| EuroEval English Summarization - CNN Dailymail | 41.66 | chrF3++ (%) | 49.7 |
| EuroEval Spanish Summarization - Mlsum ES | 24.16 | chrF3++ (%) | 47.6 |
| Open LLM Leaderboard - IFEval | 38.42 | Score | 37.7 |
| EuroEval Serbian Summarization - LR SUM SR | 21.83 | chrF3++ (%) | 37.4 |
| EuroEval Norwegian Knowledge - Idioms NO | 0.95 | MCC (x100) | 35.5 |
| EuroEval Icelandic NLU - ScaLA IS | 0.6 | Matthews correlation coefficient (%) | 31.7 |
| EuroEval Icelandic Knowledge | 3.2 | Knowledge Average Score (%) | 30.6 |
| EuroEval Faroese NLU - ScaLA FO | 0.67 | Matthews correlation coefficient (%) | 30 |
Interactive version: theaggregate.ai/model?slug=smollm2-360m-instruct · How It Works · Data refreshed daily, snapshot 2026-10-15.