Leaderboard / Alibaba (Qwen)

Qwen3-Next 80B-A3B Thinking

Name: Qwen3-Next 80B-A3B Thinking
Author: Alibaba (Qwen)

Alibaba (Qwen) · Qwen3-Next · released 2025-09-11 · data checked 2026-11-01

Qwen3-Next 80B-A3B Thinking is a 80B-parameter Mixture-of-Experts model (3B active per token) from Alibaba (Qwen) (China), released 2025-09-11 under the Apache-2.0 license. It scores 63.3 on the Small LM Index, #3 of 6 large models. It needs about 50.8 GB of memory at 4-bit with an 8K-token context.

Small LM Index

63.3

#3 of 6 scored large models

Likely 59.3–67.1 · from 4 of 6 core benchmarks

Parameters

80B

3B active · MoE

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

262K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best large model on that benchmark. Sources: 5 reported by Alibaba (Qwen). How we label sources.

MMLU-Pro Lab-reported Knowledge82.7

GPQA Diamond Lab-reported Knowledge77.2

Class best: Nemotron 3.5 Bolt 74B-A9B 88.8

AIME Lab-reported Math · AIME 202587.8

Class best: Kolibri-1 96.0

LiveCodeBench Lab-reported Coding · LiveCodeBench v668.7

Class best: Kolibri-1 85.9

IFBench Lab-reported Instruction following61.5

Class best: Kolibri-1 78.1

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 13

SWE-bench Verified≈ 51

Specifications

Lab: Alibaba (Qwen) (CN · China)
Architecture: Mixture-of-Experts
Total parameters: 80B
Active parameters: 3B
Size class: Large
Context window: 262K tokens
Input modalities: text
Reasoning: Reasoning (thinking)
License: Apache-2.0
Openness: Open weights
Memory to run: ~50.8 GB at 4-bit, 8K context (more sizes)
Hugging Face: 43.8K downloads (30 days) · 503 likes

Weights on Hugging Face ↗

Compare with

Qwen3-Next 80B-A3B Thinking vs Kolibri-1 (78.1B, index 68.3)
Qwen3-Next 80B-A3B Thinking vs Nemotron 3.5 Bolt 74B-A9B (74B, index 73.3)
Qwen3-Next 80B-A3B Thinking vs DeepSeek-R1-Distill-Llama-70B (70.6B, index 50.1)
Qwen3-Next 80B-A3B Thinking vs Llama 3.3 70B (70.6B, index 40.8)

Sources

https://huggingface.co/Qwen/Qwen3.5-9B

Something wrong here? Report it.

Run it locally

Uses unsloth/Qwen3-Next-80B-A3B-Thinking-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ4_NL, IQ4_XS, Q2_K, Q2_K_L, Q3_K_M, Q3_K_S, Q4_0….

Ollama

ollama run hf.co/unsloth/Qwen3-Next-80B-A3B-Thinking-GGUF:Q4_K_M

llama.cpp

llama-server -hf unsloth/Qwen3-Next-80B-A3B-Thinking-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit

vLLM (GPU server)

vllm serve Qwen/Qwen3-Next-80B-A3B-Thinking

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	256K context
16-bit F16	162 GB	163 GB	169 GB
8-bit Q8_0	87 GB	88 GB	94 GB
4-bit Q4_K_M	51 GB	51 GB	57 GB

Check it against your GPU, Mac or phone →