MobileLLM-R1-950M-base: benchmark results
Meta MobileLLM-R1 950M Base is the pretrained checkpoint for downstream adaptation, separate from its final reasoning SFT release. The publisher specifies 22 layers, 24 attention heads, 6 key/value heads and shared input/output embeddings. Provider: Meta. Released 2025-09-12. Access: Open.
Aggregate Rating 1251 ± 32, rank #9247 of 10916 rated models, from 143 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval English NLU - SQuAD | 79.53 | Character F1 (%) | 49.8 |
| EuroEval Spanish NLU - MLQA ES | 56.02 | Character F1 (%) | 40.1 |
| EuroEval French NLU - FQuAD | 58.52 | Character F1 (%) | 35.9 |
| EuroEval Portuguese NLU - MultiWikiQA PT | 55.59 | Character F1 (%) | 31.6 |
| EuroEval Italian NLU - SQuAD IT | 48.93 | Character F1 (%) | 31.4 |
| EuroEval German NLU - Germanquad | 42.73 | Character F1 (%) | 26.9 |
| EuroEval Finnish NLU - Tydiqa FI | 44.84 | Character F1 (%) | 26.8 |
| EuroEval Dutch NLU - SQuAD NL | 47.5 | Character F1 (%) | 26.5 |
| EuroEval Icelandic Knowledge | 1.99 | Knowledge Average Score (%) | 25.8 |
| EuroEval Faroese NLU - FoQA | 14.97 | Character F1 (%) | 25.5 |
| EuroEval Icelandic NLU - ScaLA IS | 0.07 | Matthews correlation coefficient (%) | 25.1 |
| EuroEval Czech NLU - CSFD Sentiment | 29.21 | Matthews correlation coefficient (%) | 25 |
Interactive version: theaggregate.ai/model?slug=mobilellm-r1-950m-base · How It Works · Data refreshed daily, snapshot 2026-11-08.