Alibaba/Mixture of Experts

Qwen3.8 2.4T A95B

Name: Qwen3.8 2.4T A95B
Author: Alibaba

Qwen3.8 2.4T-A95B — the open-weight release behind Qwen3.8-Max, and the first Max-class Qwen published as weights. 2.4T total, 95B active (10 routed + 1 shared of 512 experts), text-only, 256K context. The hosted Max adds vision input, non-thinking mode and 1M context; these weights do not. Custom Qwen license, not Apache.

chatcodingreasoningmultilingualmathagentictool_use

2446.18B

Parameters (95B active)

256K

Context length

Benchmarks

Quantizations

HF downloads

Architecture

MoE

Released

2026-08-12

Layers

KV Heads

Head Dim

256

Family

qwen

Quantization Options

Context length:

Quant	Bits	VRAM @ 16K	Quality
IQ2_XXS	2.38	729.3 GB 728.2 + 1.1 KV	low
IQ2_M	2.93	897.5 GB 896.4 + 1.1 KV	low
Q2_K	3.16	967.8 GB 966.7 + 1.1 KV	low
IQ3_XXS	3.25	995.3 GB 994.2 + 1.1 KV	low
IQ3_XS	3.5	1071.8 GB 1070.7 + 1.1 KV	low
Q3_K_S	3.64	1114.6 GB 1113.5 + 1.1 KV	low
IQ3_M	3.76	1151.3 GB 1150.2 + 1.1 KV	low
Q3_K_M	4	1224.7 GB 1223.6 + 1.1 KV	low
Q3_K_L	4.3	1316.4 GB 1315.3 + 1.1 KV	moderate
IQ4_XS	4.46	1365.3 GB 1364.2 + 1.1 KV	moderate
Q4_K_S	4.67	1429.5 GB 1428.4 + 1.1 KV	moderate
Q4_K_M	4.89	1496.8 GB 1495.7 + 1.1 KV	good
Q5_K_S	5.57	1704.7 GB 1703.6 + 1.1 KV	good
Q5_K_M	5.7	1744.5 GB 1743.4 + 1.1 KV	good
Q6_K	6.56	2007.4 GB 2006.4 + 1.1 KV	excellent
Q8_0	8.5	2600.6 GB 2599.6 + 1.1 KV	lossless
FP16	16	4893.9 GB 4892.8 + 1.1 KV	lossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Too big for a single GPU — plan a multi-GPU deployment

Even the lightest quant needs ~729 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →

▸ READY TO RUN THIS?RENT BY THE HOUR

RENT A GPU AND RUN QWEN3.8 2.4T A95B NOW

Rent on Vast.ai →Or RunPod →

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (6)

GPQA Diamond93.5

AA Long Context75.3

AA Coding71.9

AA Intelligence57.7

SciCode51.6

HLE42.4

Run this model

▸Easiest way to get started·Beginners

DOCS ↗

curl -fsSL https://ollama.com/install.sh | sh

$ollama run qwen:2446b-q4_K_M

Tag may need adjustment — check ollama.com/library/qwen for available tags.

▸ SETUP GUIDE

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more

Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines

HuggingFace GGUF Downloads Build Hardware

Find the best GPU for Qwen3.8 2.4T A95B

Build Hardware for Qwen3.8 2.4T A95B

QUANT	BPW	VRAM	QUALITY
IQ2_XXS	2.38	728.2 GB	65%
IQ2_M	2.93	896.4 GB	75%
Q2_K	3.16	966.7 GB	78%
IQ3_XXS	3.25	994.2 GB	82%
IQ3_XS	3.5	1070.7 GB	84%
Q3_K_S	3.64	1113.5 GB	85%
IQ3_M	3.76	1150.2 GB	86%
Q3_K_M	4	1223.6 GB	88%
Q3_K_L	4.3	1315.3 GB	90%
IQ4_XS	4.46	1364.2 GB	92%
Q4_K_S	4.67	1428.4 GB	93%
Q4_K_M	4.89	1495.7 GB	94%
Q5_K_S	5.57	1703.6 GB	96%
Q5_K_M	5.7	1743.4 GB	96%
Q6_K	6.56	2006.4 GB	97%
Q8_0	8.5	2599.6 GB	100%
FP16	16	4892.8 GB	100%