Qwen3-Next-80B-A3B-Instruct is the first in the Qwen3-Next series, featuring groundbreaking architectural innovations. It uses Hybrid Attention combining Gated DeltaNet and Gated Attention for efficient ultra-long context modeling, High-Sparsity MoE with 512 experts (10 activated + 1 shared) achieving extreme low activation ratio, and Multi-Token Prediction for improved performance and faster inference. With 80B total parameters and only 3B activated, it outperforms Qwen3-32B-Base with 10% training cost and 10x throughput for 32K+ contexts. The model performs on par with Qwen3-235B-A22B-Instruct-2507 while excelling at ultra-long-context tasks up to 256K tokens (extensible to 1M with YaRN). Architecture: 48 layers, 15T training tokens, hybrid layout of 12*(3*(Gated DeltaNet->MoE)->(Gated Attention->MoE)).
Benchmarks
评测成绩
| 评测基准 | 类别 | 分数 | 来源 |
|---|---|---|---|
| MMLU-Redux | language reasoning math general | 90.9 | 来源 |
| MultiPL-E | general language | 87.8 | 来源 |
| IFEval | general | 87.6 | 来源 |
| WritingBench | writing creativity communication | 87.3 | 来源 |
| Creative Writing v3 | creativity writing | 85.3 | 来源 |
| Arena-Hard v2 | general reasoning creativity | 82.7 | 来源 |
| MMLU-Pro | language reasoning math general | 80.6 | 来源 |
| INCLUDE | general | 78.9 | 来源 |
| MMLU-ProX | language reasoning math general | 76.7 | 来源 |
| LiveBench 20241125 | math reasoning general | 75.8 | 来源 |
| MultiIF | reasoning communication language | 75.8 | 来源 |
| GPQA | reasoning general | 72.9 | 来源 |
| BFCL-v3 | general reasoning | 70.3 | 来源 |
| AIME 2025 | math reasoning | 69.5 | 来源 |
| TAU1-Retail | reasoning communication | 60.9 | 来源 |
| SuperGPQA | reasoning general math legal healthcare finance chemistry economics physics | 58.8 | 来源 |
| TAU2-Retail | communication reasoning | 57.3 | 来源 |
| LiveCodeBench v6 | reasoning general | 56.6 | 来源 |
| HMMT25 | math | 54.1 | 来源 |
| Aider-Polyglot | general code | 49.8 | 来源 |
| PolyMATH | math reasoning spatial_reasoning multimodal vision | 45.9 | 来源 |
| TAU2-Airline | reasoning communication | 45.5 | 来源 |
| TAU1-Airline | reasoning communication | 44.0 | 来源 |
| TAU2-Telecom | communication reasoning | 13.2 | 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 |
|---|---|---|---|---|---|---|---|---|
| Helicone | $0.14 | $1.40 | 262K | — | — | ✓ | ✗ | ✗ |
| Alibaba (China) | $0.14 | $0.57 | 131K | — | — | ✓ | ✗ | ✗ |
| LLM Gateway | $0.15 | $1.20 | 131K | — | — | ✓ | ✗ | ✗ |
| Neon | $0.15 | $1.20 | 131K | — | — | ✓ | ✗ | ✗ |
| Alibaba | $0.50 | $2.00 | 131K | — | — | ✓ | ✗ | ✗ |
价格单位：美元/百万 token，数据来自社区整理，仅供参考。