K2S3-Mistral-7bx2-48layers_v1.2: benchmark results
K2S3 release published by Changgil: a 48-layer passthrough merge using layer ranges 0–24 and 8–32 from K2S3-Mistral-7b-v1.2. This depth-expanded configuration remains distinct from its 32-layer source checkpoint; the release name is not a verified learned parameter count. Provider: Changgil. Access: Open.
Aggregate Rating 1366 ± 33, rank #7271 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 49.81 | Accuracy (%) (5-shot) | 62.4 |
| Open LLM Leaderboard v1 - MMLU | 57.91 | Accuracy (%) (5-shot) | 42.1 |
| Open LLM Leaderboard v1 - WinoGrande | 74.19 | Accuracy (%) (5-shot) | 35.4 |
| Open LLM Leaderboard v1 - ARC Challenge | 56.31 | Normalized accuracy (%) (25-shot) | 35.2 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 46.12 | MC2 (%) (0-shot) | 33.5 |
| Open LLM Leaderboard v1 - HellaSwag | 77.83 | Normalized accuracy (%) (10-shot) | 32 |
Interactive version: theaggregate.ai/model?slug=k2s3-mistral-7bx2-48layers-v1-2 · How It Works · Data refreshed daily, snapshot 2026-11-08.