Nous-Capybara-7B: benchmark results
Nous Research's Capybara 7B V1 instruction fine-tune of Llama 2 7B, using the Capybara/Amplify-Instruct data project. It is distinct from the later Yi-based 34B release. Provider: Nous Research. Access: Open.
Aggregate Rating 1309 ± 26, rank #8563 of 10916 rated models, from 15 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 51.13 | MC2 (%) (0-shot) | 50.8 |
| Open LLM Leaderboard v1 - HellaSwag | 80.73 | Normalized accuracy (%) (10-shot) | 42.1 |
| Open Korean LLM - Ko-IFEval | 26.17 | Strict accuracy, prompt/instruction mean (%) | 38.3 |
| Open Korean LLM - KorNAT-Harmless | 61.9 | Normalized accuracy (%) | 35.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 55.29 | Normalized accuracy (%) (25-shot) | 33.1 |
| Open LLM Leaderboard v1 - WinoGrande | 73.32 | Accuracy (%) (5-shot) | 31.7 |
| Open Korean LLM - Ko-GPQA-Diamond | 21.72 | Normalized accuracy (%) | 29.4 |
| Open LLM Leaderboard v1 - MMLU | 48.72 | Accuracy (%) (5-shot) | 28 |
| Open LLM Leaderboard v1 - GSM8K | 6.97 | Accuracy (%) (5-shot) | 26.2 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 5.76 | Exact match, flexible extract (%) | 25.3 |
| Open Korean LLM - Ko-EQ-Bench | 0 | EQ-Bench score | 24.5 |
| Open Korean LLM Leaderboard | 30.3 | Average Score (%) | 18 |
Interactive version: theaggregate.ai/model?slug=nous-capybara-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.