dolphin-2.9.4-llama3.1-8B: benchmark results
Cognitive Computations' Dolphin 2.9.4, a fine-tune of Meta's Llama 3.1 8B using ChatML formatting, trained for multilingual instruction following, function calling and system-prompt adherence. Provider: Cognitive Computations. Released 2024-08-04. Access: Open.
Aggregate Rating 1266 ± 19, rank #1529 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 24.63 | Strict accuracy, prompt/instruction mean (%) | 30.7 |
| Open Portuguese LLM - OAB Exams | 30.48 | Accuracy (%) | 26.1 |
| Open Korean LLM - Ko-EQ-Bench | 0 | EQ-Bench score | 24.5 |
| Open LLM Leaderboard - IFEval | 27.57 | Score | 24.5 |
| Open Portuguese LLM - ASSIN2 RTE | 68.28 | Macro F1 (%) | 24.1 |
| Open Portuguese LLM - BLUEX | 27.4 | Accuracy (%) | 23.5 |
| Open Portuguese LLM - ENEM | 24.49 | Accuracy (%) | 20.9 |
| Open LLM Leaderboard - BBH | 35.24 | Score | 19.6 |
| Open LLM Leaderboard - GPQA | 26.34 | Score | 17.3 |
| Open Portuguese LLM - FaQuAD NLI | 43.97 | Macro F1 (%) | 17.1 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 0.83 | Exact match, flexible extract (%) | 10.8 |
| Open Korean LLM - KorNAT-Helpful | 41.18 | Normalized accuracy (%) | 10.8 |
Interactive version: theaggregate.ai/model?slug=dolphin-2-9-4-llama3-1-8b · How It Works · Data refreshed daily, snapshot 2026-10-23.