LMUnit-qwen2.5-72B: benchmark results
Contextual AI LMUnit is a specialized natural-language unit-test evaluator fine-tuned from Qwen/Qwen2.5-72B-Instruct. Given a prompt, response and unit test, it produces a score from 1 to 5 for criterion satisfaction; it remains distinct from the parent instruction model and general-purpose chat releases. Provider: Contextual AI. Access: Open.
Aggregate Rating 1662 ± 43, rank #542 of 10916 rated models, from 18 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| RewardBench 2 Factuality | 87.16 | Accuracy (%) | 99.6 |
| RewardBench 2 | 82.08 | Score (%) | 99.5 |
| RewardBench 2 Ties | 90.14 | Score (%) | 99.5 |
| RewardBench 2 Focus | 96.77 | Score (%) | 98.8 |
| RewardBench 2 Safety | 91.33 | Score (%) | 89.1 |
| RewardBench 2 Precise IF | 54.37 | Score (%) | 86.9 |
| RewardBench 2 Math | 72.68 | Score (%) | 81.2 |
| FormalRewardBench - Pointwise | 41.2 | Pointwise accuracy (%): the model scores the correct and the | 56 |
| IF-RewardBench - Overall Assessment | 0.08 | Kendall tau-b (-1 to 1) between the judge ranking of the res | 43.8 |
| IF-RewardBench - Overall Assessment (Single-Turn) | 0.13 | Kendall tau-b (-1 to 1) between the judge ranking of the res | 43.8 |
| IF-RewardBench - Overall Assessment (Multi-Turn) | 0.11 | Kendall tau-b (-1 to 1) between the judge ranking of the res | 37.5 |
| IF-RewardBench - Overall Assessment (System-Prompt) | 0.01 | Kendall tau-b (-1 to 1) between the judge ranking of the res | 37.5 |
Interactive version: theaggregate.ai/model?slug=lmunit-qwen2-5-72b · How It Works · Data refreshed daily, snapshot 2026-11-08.