Llama-3-Taiwan-70B-Instruct: benchmark results
Provider: Meta.
Aggregate Rating 1604 ± 43, rank #973 of 10916 rated models, from 9 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Twinkle Eval - TW-Legal Benchmark v1 | 59.78 | Accuracy (%) | 96.4 |
| Twinkle Eval - TMMLU+ | 69.64 | Accuracy (%) | 75 |
| TW-LegalBench - Essay Questions | 55 | Rubric points judged fully correct (out of 580) across the 1 | 66.7 |
| TW-LegalBench - Administrative Law | 61 | Administrative-law questions: accuracy (%) on four-option si | 58.3 |
| TW-LegalBench - Civil Law | 60.2 | Civil-law questions: accuracy (%) on four-option single-answ | 58.3 |
| TW-LegalBench - Criminal Law | 65.2 | Criminal-law questions: accuracy (%) on four-option single-a | 58.3 |
| TW-LegalBench - Constitutional Law | 71.5 | Constitutional-law questions: accuracy (%) on four-option si | 50 |
| TW-LegalBench - International Law | 64.5 | International-law questions: accuracy (%) on four-option sin | 50 |
| Twinkle Eval - MMLU | 41.09 | Accuracy (%) | 10.7 |
Interactive version: theaggregate.ai/model?slug=llama-3-taiwan-70b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.