DeepSeek LLM Chat 67B: benchmark results
DeepSeek's first-gen 67B chat model, instruction-tuned from a base trained on 2T English/Chinese tokens; open weights allowing commercial use. Provider: DeepSeek. Released 2023-11-29. Access: Open.
Aggregate Rating 1461 ± 16, rank #3285 of 10916 rated models, from 505 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - MuSR | 50.59 | Score | 98.9 |
| ChineseSafe Benchmark | 76.76 | Accuracy (%) | 98.1 |
| HELM MMLU - Business Ethics | 86 | EM | 96.8 |
| HELM Lite - LegalBench - Function of Decision Section | 44.96 | Exact Match (%) | 89.4 |
| HELM AIR-Bench 2024 - #6.23: Medical/Pharmaceutical | 23.33 | Refusal Rate (%) | 85.5 |
| HELM AIR-Bench 2024 - #6.25: Financial | 16.67 | Refusal Rate (%) | 84.9 |
| HELM AIR-Bench 2024 - #39-40.36: Characterization of identity - Disability | 100 | Refusal Rate (%) | 83.7 |
| HELM AIR-Bench 2024 - #6.26: Government services | 6.67 | Refusal Rate (%) | 83.1 |
| HELM MMLU - Human Aging | 82.51 | EM | 82.7 |
| HELM AIR-Bench 2024 - #39-40.55: Classification of individuals - Health conditions | 100 | Refusal Rate (%) | 82.6 |
| HELM AIR-Bench 2024 - #6.22: Legal | 6.67 | Refusal Rate (%) | 82.6 |
| HELM AIR-Bench 2024 - #4.4: Employment | 23.33 | Refusal Rate (%) | 82 |
Interactive version: theaggregate.ai/model?slug=deepseek-llm-chat-67b · How It Works · Data refreshed daily, snapshot 2026-11-08.