K2S3-Mistral-7B-v1.2: benchmark results
K2S3 release v1.2 published by Changgil: a full-parameter supervised fine-tune of mistralai/Mistral-7B-v0.1, using alpaca-gpt4-data and OpenOrca samples. The release configuration uses 32 Mistral decoder layers; versioned siblings are not interchangeable checkpoints. Provider: Changgil. Access: Open.
Aggregate Rating 1373 ± 33, rank #7090 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 55.04 | Accuracy (%) (5-shot) | 66.7 |
| Open LLM Leaderboard v1 - MMLU | 58.04 | Accuracy (%) (5-shot) | 42.4 |
| Open LLM Leaderboard v1 - WinoGrande | 75.37 | Accuracy (%) (5-shot) | 41.8 |
| Open LLM Leaderboard v1 - HellaSwag | 79.61 | Normalized accuracy (%) (10-shot) | 38.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 57.51 | Normalized accuracy (%) (25-shot) | 37.7 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 46.7 | MC2 (%) (0-shot) | 35.2 |
Interactive version: theaggregate.ai/model?slug=k2s3-mistral-7b-v1-2 · How It Works · Data refreshed daily, snapshot 2026-11-08.