qwen3.5:27b

qwen35 · 27.8B · Q4_K_M

THINKING MODEL ECO MODE

AZW GTR Pro (AMD RYZEN AI MAX+ 395)

125 GB · Ubuntu 24.04.4 LTS

Tested on March 5, 2026

Top 99% Compare

Global Score

20 /100

Not Rec.

Hardware Fit

37/100

Quality

12/100

Get this model

🦙

Ollama

ollama pull qwen3.5:27b

View on Ollama Library

ollama.com/library/qwen3.5

Get it in LM Studio

Search and download models directly from the app

🤗

Find on HuggingFace

GGUF versions & quantizations

Hardware

Machine: AZW GTR Pro
CPU: AMD RYZEN AI MAX+ 395
Cores: 32 threads (16 cores)
Frequency: 3 GHz
RAM: 125 GB
GPU: AMD Radeon 8060S
OS: Ubuntu 24.04.4 LTS
Arch: x64
Power Mode: low-power

Performance

Tokens/sec Tokens generated per second — higher is better: 10.6
Standard deviation Variation of speed across benchmark runs — lower means more consistent: ±0.0
First chunk latency Delay before the first streamed chunk arrives from runtime: 358 ms
Time to first token How fast the model starts responding — lower is better: 13.7 s
Load time Time to load the model into memory before inference: 10.8 s
Memory usage RAM consumed during inference — percentage of total system memory: 39.7 GB (32%)
Total tokens Total number of tokens generated across all benchmark prompts: 1429
Thinking tokens (est.) Estimated internal reasoning tokens — models with "thinking" generate more tokens: ~699

Score breakdown

Speed

9/50

Time to first token

0/20

Memory

28/30

Quality

Reasoning

2/20

Coding

0/20

Instruction following

2/20

Structured output

0/15

Math

0/15

Multilingual

8/10

Category levels

Reasoning: Poor Coding: Poor Instruction Following: Poor Structured Output: Poor Math: Poor Multilingual: Strong

Metadata

Spec version: 0.2.1
Runtime: Ollama 0.17.4
Model format: GGUF
Hardware profile: HIGH-END
Result hash: 8705f207e39cd5190393de12503bbe7fbcc0afa77ca639d8cb1233d2cf0b4a6b

Interpretation

Hardware fit: 37/100. Overall suitability: NOT RECOMMENDED (Global 20/100). Category profile: Reasoning: Poor, Coding: Poor, Instruction Following: Poor, Structured Output: Poor, Math: Poor, Multilingual: Strong. Warning: model produced very low accuracy on quality tasks — results may be unusable despite good hardware performance.

Warnings

Model produced very low accuracy on quality tasks — results may be unusable despite good hardware performance.
System was in low-power mode during this benchmark.
CPU appears throttled (2.2 GHz current vs 3.0 GHz nominal, 73%).

Disqualifiers

Time to first token too high: 13722ms (maximum: 10000ms for HIGH-END profile)

Bench Environment

Thermal: nominal CPU load: avg 3% (peak 4%)

Run yours now

$ npm install -g metrillm@latest

$ metrillm

Requires Node 20+ and Ollama or LM Studio running

Or run without installing: npx metrillm@latest