merlinite-7B: benchmark results
IBM Research's Merlinite-7b, a Mistral 7B derivative trained with their LAB (Large-scale Alignment for chatBots) synthetic-data method, using Mixtral-8x7B-Instruct as the teacher model. Provider: IBM. Released 2024-03-02. Access: Open.
Aggregate Rating 1398 ± 27, rank #1232 of 1656 rated models, from 18 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open CoT - LSAT Analytical Reasoning | 10.43 | CoT Gain (%) | 95.4 |
| Open LLM Leaderboard - MuSR | 44.12 | Score | 78.7 |
| Open LLM Leaderboard v1 - MMLU | 64.91 | Accuracy (%) (5-shot) | 78.4 |
| Open LLM Leaderboard v1 - WinoGrande | 79.72 | Accuracy (%) (5-shot) | 70.7 |
| Open LLM Leaderboard v1 - HellaSwag | 84.52 | Normalized accuracy (%) (10-shot) | 66.6 |
| Open LLM Leaderboard v1 - ARC Challenge | 63.65 | Normalized accuracy (%) (25-shot) | 63.7 |
| Open LLM Leaderboard v1 - GSM8K | 41.09 | Accuracy (%) (5-shot) | 56.2 |
| Open LLM Leaderboard - GPQA | 29.7 | Score | 52.6 |
| Open LLM Leaderboard - BBH | 50.07 | Score | 48.5 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 50.15 | MC2 (%) (0-shot) | 47.3 |
| Open CoT - LogiQA | 3.83 | CoT Gain (%) | 46.2 |
| Open CoT Leaderboard | 5.8 | Average CoT Gain (%) | 39.7 |
Interactive version: theaggregate.ai/model?slug=merlinite-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.