MegaBeam-Mistral-7B-300k: benchmark results
Amazon's 7B long-context fine-tune of Mistral-7B-Instruct-v0.2, extending support to about 320,000 tokens of input context, deployable via vLLM or Amazon SageMaker. Provider: Amazon. Released 2024-05-13. Access: Open.
Aggregate Rating 1364 ± 17, rank #1356 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-GPQA-Diamond | 26.77 | Normalized accuracy (%) | 88.9 |
| Open Korean LLM - KorNAT-Helpful | 50.98 | Normalized accuracy (%) | 88.3 |
| Open Korean LLM - KorNAT-Harmless | 65.58 | Normalized accuracy (%) | 81.5 |
| Open Chinese LLM - TruthfulQA MC | 57.78 | Accuracy (%) | 74 |
| Open Korean LLM - Ko-IFEval | 33.97 | Strict accuracy, prompt/instruction mean (%) | 69.5 |
| Open Korean LLM - Ko-EQ-Bench | 17.79 | EQ-Bench score | 63.9 |
| Open LLM Leaderboard - IFEval | 52.03 | Score | 62 |
| Open Korean LLM Leaderboard | 36.72 | Average Score (%) | 58.6 |
| Open Ko-LLM Leaderboard | 36.72 | Average (%) | 58.5 |
| Open Portuguese LLM - ASSIN2 RTE | 89.87 | Macro F1 (%) | 52.8 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 15.85 | Exact match, flexible extract (%) | 45.7 |
| Open Portuguese LLM - FaQuAD NLI | 61.24 | Macro F1 (%) | 45.3 |
Interactive version: theaggregate.ai/model?slug=megabeam-mistral-7b-300k · How It Works · Data refreshed daily, snapshot 2026-10-20.