← All models

Qwen 3.5 (27B, Q4)

generalApache 2.0tools

Alibaba's dense Qwen 3.5 model at 27B parameters. Multimodal (text + image + video), native tool calling, and 262K native context. A capable on-device option for machines with a 24GB+ inference budget.

At a glance

Parameters
27B
Quantization
Q4_K_M
Context window
262K tokens
Approx. size
17.1 GB
Engines
llama.cpp, MLX, Ollama
License
Apache-2.0 (open)
Version
v1.0.0 , released 2026-09-04
Family
qwen

Tuning

Defaults a gezel client applies out of the box, from real evaluation runs against this exact quantization.

sampling.temperature
0.7
sampling.topP
0.8
sampling.topK
20
sampling.minP
0
sampling.maxTokens
16384
sampling.repetitionPenalty
1.1
sampling.repetitionContext
20
reasoning.thinkingBudget
2048
reasoning.enableThinking
true
profiles
thinking-general, thinking-coding, thinking-precise, instruct, creative

Model behaviors

Client-side behavior modules the catalog enables for this model (reasoning-tag handling, fabrication detection, prompt shaping).

Sources

llama.cpp (GGUF)
unsloth/Qwen3.5-27B-MTP-GGUF · Qwen3.5-27B-Q4_K_M.gguf · Q4_K_M
MLX
mlx-community/Qwen3.5-27B-4bit
Ollama
qwen3.5:27b
Upstream
https://huggingface.co/Qwen/Qwen3.5-27B
Manifest
View on GitHub

Tags