A powerful Mixture-of-Experts (MoE) language model with 671B total parameters (37B activated per token). Features Multi-head Latent Attention (MLA), auxiliary-loss-free load balancing, and multi-token prediction training. Pre-trained on 14.8T tokens with strong performance in reasoning, math, and code tasks.
发布日期2024年12月25日
参数规模671B
上下文长度128K
许可证MIT + Model License (Commercial use allowed)
知识截止—
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| DROP | reasoning math | 91.6 | 来源 |
| CLUEWSC | language reasoning | 90.9 | 来源 |
| MATH-500 | math reasoning | 90.2 | 来源 |
| MMLU-Redux | language reasoning math general | 89.1 | 来源 |
| MMLU | general reasoning language math | 88.5 | 来源 |
| C-Eval | general reasoning | 86.5 | 来源 |
| IFEval | general | 86.1 | 来源 |
| HumanEval-Mul | reasoning | 82.6 | 来源 |
| Aider-Polyglot Edit | general code | 79.7 | 来源 |
| MMLU-Pro | language reasoning math general | 75.9 | 来源 |
| FRAMES | reasoning search | 73.3 | 来源 |
| CSimpleQA | general language | 64.8 | 来源 |
| GPQA | reasoning general | 59.1 | 来源 |
| Aider-Polyglot | general code | 49.6 | 来源 |
| LongBench v2 | long_context reasoning general | 48.7 | 来源 |
| CNMO 2024 | math | 43.2 | 来源 |
| SWE-Bench Verified | reasoning frontend_development code | 42.0 | 来源 |
| AIME 2024 | math reasoning | 39.2 | 来源 |
| LiveCodeBench | reasoning general code | 37.6 | 来源 |
| SimpleQA | general reasoning | 24.9 | 来源 |
Pricing
API 价格对比
| 服务商 | 输入价 | 输出价 | 上下文 | 吞吐(tok/s) | 延迟(s) | 函数调用 | 代码执行 | 联网搜索 |
|---|---|---|---|---|---|---|---|---|
| iFlow | $0.00 | $0.00 | 128K | — | — | ✓ | ✗ | ✗ |
| Alibaba (China) | $0.29 | $1.15 | 66K | — | — | ✓ | ✗ | ✗ |
| Helicone | $0.56 | $1.68 | 128K | — | — | ✓ | ✗ | ✗ |
价格单位：美元/百万 token，数据来自社区整理，仅供参考。