Benchmark result

Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp on 2× NVIDIA GeForce RTX 5060 — 40.0 tok/s

Measured with llama.cpp b10884-434ddbbc0 on September 10, 2026.

What the model built

The coding scenario asks for a playable game in a single HTML file. This is exactly what the model returned, unedited; the frame only adds a line that tells this page how tall it is.

How this run compares

1st fastest of 2 runs of this model on 2× NVIDIA GeForce RTX 5060 · Multi-GPU · median 39.4 tok/s.

sort

140.0tok/s82.2llama.cppthis run
238.8tok/s85.2llama.cpp

Compare the nearest 5 →

Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp on other hardware

Reproduce this run

toolllama.cpp b10884-434ddbbc0

modelQwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp

context131,072

thinkingon

effortxhigh (model default)

samplingtemperature 0.6000000238418579 · top_p 0.8999999761581421 · top_k 0 · min_p 0.05000000074505806 · presence_penalty 0 · repeat_penalty 1

clientv0.4.59+97

Set those in llama.cpp, then:

llm-benchmark benchmark --model "Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp" --tool "llama.cpp"

The client prompts for context length and thinking mode, and for the KV cache dtype on Unsloth Studio. Sampling is left at the model default — the values above are what the tool reported using, not overrides the client sent.