stablelm-2-12B-chat: benchmark results
Stability AI Stable LM 2 12B Chat is an instruction and preference tuned decoder-only language model, using public and synthetic datasets with direct preference optimization. It remains distinct from the pretrained Stable LM 2 12B base checkpoint. Provider: Stability AI. Released 2024-04-08. Access: Open.
Aggregate Rating 1422 ± 26, rank #1138 of 1656 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 62.01 | MC2 (%) (0-shot) | 78.5 |
| RewardBench Chat | 96.65 | Accuracy (%) | 77.8 |
| Open LLM Leaderboard v1 - HellaSwag | 85.96 | Normalized accuracy (%) (10-shot) | 77.7 |
| Open Portuguese LLM - ASSIN2 RTE | 92.19 | Macro F1 (%) | 75.7 |
| RewardBench Reasoning | 89.45 | Accuracy (%) | 70.4 |
| Open LLM Leaderboard v1 - GSM8K | 57.85 | Accuracy (%) (5-shot) | 69.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 64.85 | Normalized accuracy (%) (25-shot) | 68.1 |
| Open LLM Leaderboard v1 - WinoGrande | 78.53 | Accuracy (%) (5-shot) | 63.8 |
| Open Portuguese LLM - OAB Exams | 44.69 | Accuracy (%) | 59 |
| Open Portuguese LLM - FaQuAD NLI | 67.71 | Macro F1 (%) | 55 |
| RewardBench | 76.42 | Score (%) | 52.8 |
| Open LLM Leaderboard v1 - MMLU | 61.06 | Accuracy (%) (5-shot) | 50.7 |
Interactive version: theaggregate.ai/model?slug=stablelm-2-12b-chat · How It Works · Data refreshed daily, snapshot 2026-10-20.