K2S3-Mistral-7B-v1.3: benchmark results
K2S3 release v1.3 published by Changgil: a full-parameter supervised fine-tune of Changgil/K2S3-Mistral-7b-v1.2, 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 1377 ± 33, rank #6928 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 54.89 | Accuracy (%) (5-shot) | 66.6 |
| Open LLM Leaderboard v1 - WinoGrande | 77.03 | Accuracy (%) (5-shot) | 52.7 |
| Open LLM Leaderboard v1 - MMLU | 58.44 | Accuracy (%) (5-shot) | 43.4 |
| Open LLM Leaderboard v1 - ARC Challenge | 58.11 | Normalized accuracy (%) (25-shot) | 39 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 47.21 | MC2 (%) (0-shot) | 36.8 |
| Open LLM Leaderboard v1 - HellaSwag | 78.94 | Normalized accuracy (%) (10-shot) | 36.3 |
Interactive version: theaggregate.ai/model?slug=k2s3-mistral-7b-v1-3 · How It Works · Data refreshed daily, snapshot 2026-11-08.