LLMOpen weights
Qwen3-Next-80B-A3B
Alibaba
Updated May 22, 2026 · Benchmarks via Artificial Analysis, specs & pricing via OpenRouter · Methodology · Report an error
Specifications
- Type
- LLM
- Access
- Open weights
- Released
- September 10, 2025
- License
- Apache 2.0
- Parameters
- 80B (3B active)
- Context window
- 262,144 tokens
- Output speed
- 147 tok/s
- Latency (TTFT)
- 1.14s
- Input
- Text
- Output
- Text
- API pricing
- $0.50 in · $6.00 out / 1M tokens
Benchmarks
Reasoning
| GPQA Diamond | 75.9 |
Coding
| LiveCodeBench | 78.4 |
Math
| AIME 2025 | 84.3 |
General
| MMLU-Pro | 82.4 |
| Humanity’s Last Exam | 11.7 |
Our take
An ultra-sparse hybrid architecture (gated DeltaNet + gated attention, 512 experts) delivering large throughput gains at ~3B active params — the architectural preview for Qwen3.5.