Trinity-2-Codestral-22B: benchmark results
Migel Tissera's Trinity-2 coding model, a fine-tune of Mistral's Codestral 22B with compute sponsored by KindoAI. Provider: migtissera. Released 2024-08-07. Access: Open.
Aggregate Rating 1493 ± 19, rank #811 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 55.16 | Strict accuracy, prompt/instruction mean (%) | 96.6 |
| Open Ko-LLM Leaderboard | 47.26 | Average (%) | 94.7 |
| Open Korean LLM Leaderboard | 47.26 | Average Score (%) | 94.7 |
| Open Portuguese LLM - FaQuAD NLI | 82.13 | Macro F1 (%) | 93.8 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 57.39 | Exact match, flexible extract (%) | 91.5 |
| Open Korean LLM - Ko-EQ-Bench | 38.15 | EQ-Bench score | 89.3 |
| Open LLM Leaderboard - BBH | 55.93 | Score | 76.8 |
| Open Korean LLM - KorNAT-CKA | 28.16 | Normalized accuracy (%) | 74.4 |
| Open LLM Leaderboard - GPQA | 31.46 | Score | 71.4 |
| Open Portuguese LLM - BLUEX | 57.44 | Accuracy (%) | 70 |
| Open Korean LLM - Ko-Winogrande | 62.67 | Accuracy (%) | 69.8 |
| Open Korean LLM - KorNAT-Helpful | 48.45 | Normalized accuracy (%) | 68.1 |
Interactive version: theaggregate.ai/model?slug=trinity-2-codestral-22b · How It Works · Data refreshed daily, snapshot 2026-10-23.