K2S3-Mistral-7B-v1.4: benchmark results
K2S3 release v1.4 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 1400 ± 33, rank #6119 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - WinoGrande | 80.35 | Accuracy (%) (5-shot) | 73.5 |
| Open LLM Leaderboard v1 - GSM8K | 42.84 | Accuracy (%) (5-shot) | 57.2 |
| Open LLM Leaderboard v1 - MMLU | 62.1 | Accuracy (%) (5-shot) | 54 |
| Open LLM Leaderboard v1 - HellaSwag | 82.33 | Normalized accuracy (%) (10-shot) | 50.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 60.15 | Normalized accuracy (%) (25-shot) | 46.6 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 48.6 | MC2 (%) (0-shot) | 41.8 |
Interactive version: theaggregate.ai/model?slug=k2s3-mistral-7b-v1-4 · How It Works · Data refreshed daily, snapshot 2026-11-08.