multi_verse_model: benchmark results
MTSAIR's Mistral-7B-based conversational fine-tune, trained with the company's in-house 'new approach' training method. Provider: MTS AI. Released 2024-03-07. Access: Open.
Aggregate Rating 1432 ± 18, rank #1092 of 1656 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - KorNAT-Harmless | 77.98 | Normalized accuracy (%) | 98.3 |
| Open Korean LLM - KorNAT-CKA | 36.62 | Normalized accuracy (%) | 96.8 |
| Open Chinese LLM - TruthfulQA MC | 63.84 | Accuracy (%) | 95.1 |
| Open Portuguese LLM - FaQuAD NLI | 79.44 | Macro F1 (%) | 87.1 |
| Open Korean LLM - KorNAT-Helpful | 50 | Normalized accuracy (%) | 83.5 |
| Open Korean LLM Leaderboard | 40.94 | Average Score (%) | 79.8 |
| Open Ko-LLM Leaderboard | 40.94 | Average (%) | 79.7 |
| Open Portuguese LLM - ASSIN2 RTE | 92.26 | Macro F1 (%) | 77 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 40.49 | Exact match, flexible extract (%) | 75.1 |
| Open Korean LLM - Ko-EQ-Bench | 19.38 | EQ-Bench score | 65.7 |
| Open Korean LLM - Ko-Winogrande | 61.09 | Accuracy (%) | 59.7 |
| Open Chinese LLM - GSM8K | 47.23 | Accuracy (%) | 56.2 |
Interactive version: theaggregate.ai/model?slug=multi-verse-model · How It Works · Data refreshed daily, snapshot 2026-10-20.