rho-math-1B-v0.1: benchmark results
Microsoft's Rho-Math v0.1 base math model, pretrained with Selective Language Modeling to focus the learning loss on selected useful tokens. It is distinct from the separately released interpreter fine-tune. Provider: Microsoft. Access: Open.
Aggregate Rating 1145 ± 35, rank #10028 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 27.05 | Accuracy (%) (5-shot) | 14.5 |
| Open LLM Leaderboard v1 - WinoGrande | 59.75 | Accuracy (%) (5-shot) | 12.9 |
| Open LLM Leaderboard v1 - ARC Challenge | 34.3 | Normalized accuracy (%) (25-shot) | 12.2 |
| Open LLM Leaderboard v1 - HellaSwag | 53.34 | Normalized accuracy (%) (10-shot) | 11.3 |
| Open LLM Leaderboard v1 - GSM8K | 0 | Accuracy (%) (5-shot) | 4.5 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 35.48 | MC2 (%) (0-shot) | 2.3 |
Interactive version: theaggregate.ai/model?slug=rho-math-1b-v0-1 · How It Works · Data refreshed daily, snapshot 2026-11-08.