llmrun

Qwen 3.8 Models — Hardware Requirements

12 Qwen 3.8 models from Alibaba and the community, from the smallest that runs in 1.9 GB of VRAM at Q4 up to 2446.2B parameters. Every row links to full quantization tables and GPU compatibility.

All Qwen 3.8 Models by Size

ModelParamsContext
Qwen3.8 2B2.3B262K
Qwen3.8 27B DSpark27B262K
Qwen3.8 27B27.8B262K
Qwen3.8 Flash Next180.0B262K
Qwen3.8 2.4T A95B2446.2B262K
Community variants and fine-tunes (44)
ModelParamsContext
Qwen3.8 2B Distill2.3B262K
Qwen3.8 4B Distill4.7B262K
Qwen3.8 4B4.7B262K
Qwen3.8 27B Escha W26.3B262K
Qwen3.8 27B EXL3 3.5bpw7.7B262K
Qwen3.8 9B Distill Uncensored Heretic9.4B262K
Qwen3.8 9B9.7B262K
Qwen3.8 9B Distill9.7B262K
Instinct Python Coder Gemma4 12B Qwen3.8 27B12B—
Qwen3.8 Flash Next Tq4a Tq2e G6416.6B262K
Qwen3.8 Flash Next VQ 2.1bpw24.6B262K
Qwen3.8 27B Abliterated MTPLX Optimized Speed26.9B262K
Qwen3.8 27B Heretic Abliterated Uncensored26.9B262K
Qwen3.8 27B Obliterated E0326.9B262K
Qwen3.8 Whittle MoE 27B A17.8B26.9B262K
Qwen3.8 27B DFlash227B262K
Qwen3.8 27B Uncensored Ninfer27B—
Qwen3.8 27B EfficientThink Uncensored K3 Opus5 Grok4.6 GPT5.6Sol SFT SimPO DFlash227B—
Qwen3.8 27B ASCII Condensed27B—
DarkIdol Qwen3.8 27B V1.1 Voice27B—
Qwen3.8 27B DSpark Agentic27B262K
Qwen3.8 27B MTPLX Optimized Speed27.4B262K
Qwen3.8 27B MTPLX Optimized Quality27.4B262K
Qwen3.8 27B MTPLX Optimized Speed FP1627.4B262K
Ektome Qwen3.8 27B PristinelyUncensored27.4B262K
Grug V1.1 Qwen 3.8 27B27.4B262K
Huihui Qwen3.8 27B Abliterated27.8B262K
Qwen3.8 27B OBLITERATED Mythos Class Agentic27.8B262K
Qwen3.8 27B OBLITERATED27.8B262K
Qwen3.8 27B TURBO Fable Cold Fusion 735 882 Heretic Uncensored NM DAU27.8B262K
Qwen3.8 27B Uncensored27.8B—
Qwen3.8 27B AEON ULTIMATE UNCENSORED BF1627.8B262K
Qwen3.8 27B ABLITERATED BF1627.8B262K
Swift 1.5 Qwen3.8 27B27.8B262K
Qwen3.8 27B Cold Fusion GAIN V1.127.8B262K
Qwen3.8 27B Pi27.8B262K
Qwen3.8 Distill 35B A3B Coder Abliterated35B262K
Qwen3.8 35B A3B Distill35.1B262K
Qwen3.8 Flash Next 40B Prune Research40.7B262K
Qwen3.8 Flash Coder 85gb BF1642.6B262K
Qwen3.8 Flash Next MTPLX Optimized Speed126.2B262K
Qwen3.8 Flash Next MTPLX Bare Speed126.2B262K
Qwen3.8 Flash Next Uncensored180.0B—
Qwen3.8 Flash Next OQ5e Mtp180.0B262K

How Qwen 3.8 Compares — llmrun Composite Score

Qwen3.8 27B is the highest-scoring Qwen 3.8 model with an llmrun Composite Score of 122.1 — #11 among 100 open models. The top proprietary model, GPT 6 Astra, scores 159.0. Click a model to see its full benchmark breakdown.

GPT 6 Astra · proprietary159.0
GPT 6 Astra Max · proprietary156.4
Claude Opus 5.5 Max · proprietary153.6
GPT 6.1 Sol Max · proprietary152.5
GPT 5.6 Sol · proprietary148.7
MiMo V2.6 Pro RL133.4
Kimi K3132.8
MiMo V2.6 Flash RL131.9
GLM 5.3130.1
DeepSeek V4.1 Flash129.5
DeepSeek V4 Pro 0813129.2
DeepSeek V4 Flash 0731127.6
Muse Spark Open 816B A42B125.3
Qwen3.8 27B122.1
llmrun Composite Score, open and proprietary models (how it is built) · ■ Qwen 3.8 · ■ other open models · ■ proprietary · bars start at 120

Frequently Asked Questions

How much VRAM do I need to run a Qwen 3.8 model?
The smallest Qwen 3.8 model, Qwen3.8 2B, runs from 1.9 GB of VRAM at Q4_K_M (smaller low-bit quantizations start at 1.4 GB with a noticeable quality loss). Larger family members need proportionally more — see the table above for every model.
Which Qwen 3.8 models can I run on a 16 GB GPU?
12 of 49 Qwen 3.8 models fit in 16 GB of VRAM at Q4_K_M, including Qwen3.8 4B Distill, Qwen3.8 9B, Qwen3.8 4B.
What is the most popular Qwen 3.8 model to run locally?
Qwen3.8 27B is the most downloaded Qwen 3.8 model in local-friendly quantized formats. It runs from 17.9 GB of VRAM at Q4_K_M.
How do Qwen 3.8 models score on benchmarks?
Qwen3.8 27B leads the family with an llmrun Composite Score of 122.1, ranking #11 among 100 open models, while the top proprietary model, GPT 6 Astra, scores 159.0. See the comparison chart above for the full standings.