Benchmark result
unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_S on NVIDIA GeForce GTX 1070 with Max-Q Design +1 more — 36.4 tok/s
Measured with llama.cpp b10760-0f3a71be1 on September 4, 2026.
What the model built
Open full screen →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
2nd fastest of 2 runs of this model on NVIDIA GeForce GTX 1070 with Max-Q Design +1 more · Multi-GPU · median 40.6 tok/s.
sort
- 144.8tok/s82.4llama.cpp · Q5_K_S
- 236.4tok/s80.4llama.cpp · Q5_K_Sthis run
Reproduce this run
toolllama.cpp b10760-0f3a71be1
modelunsloth/Qwen3.8-27B-GGUF:UD-Q5_K_S (Q5_K_S)
context65,536
thinkingon
samplingtemperature 0.4000000059604645 · top_p 0.800000011920929 · top_k 20 · min_p 0 · presence_penalty 0 · repeat_penalty 1
clientv0.4.53+97
Set those in llama.cpp, then:
llm-benchmark benchmark --model "unsloth/Qwen3.8-27B-GGUF:UD-Q5_K_S" --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.