Benchmark result
unsloth/Qwen3.5-9B-MTP-GGUF on NVIDIA GeForce RTX 4070 — 102.9 tok/s
Measured with llama.cpp unsloth-studio/v0.1.810-beta on September 18, 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
4th fastest of 10 runs of this model on NVIDIA GeForce RTX 4070 · median 99.8 tok/s.
sort
- 1110.8tok/s73.3Unsloth Studio · UD-Q4_K_XL · KV f16
- 2105.3tok/s61.7Unsloth Studio · UD-Q4_K_XL · KV f16
- 3105.1tok/s71.6Unsloth Studio · UD-Q4_K_XL · KV q8_0
- 4102.9tok/s66.5llama.cppthis run
- 5102.0tok/s69.8Unsloth Studio · UD-Q5_K_XL · KV f16
- 697.7tok/s67.1Unsloth Studio · UD-Q5_K_XL · KV f16
- 795.8tok/s71.5llama.cpp
- 863.0tok/s68.2Unsloth Studio · UD-Q5_K_XL · KV q4_0
- 956.0tok/s69.7Unsloth Studio · Q6_K · KV f16
- 1017.6tok/s66.7Unsloth Studio · UD-Q4_K_XL · KV f16
Reproduce this run
toolllama.cpp unsloth-studio/v0.1.810-beta
modelunsloth/Qwen3.5-9B-MTP-GGUF
context262,144
thinkingon
samplingtemperature 0.800000011920929 · top_p 0.949999988079071 · top_k 40 · min_p 0.05000000074505806 · presence_penalty 0 · repeat_penalty 1
clientv0.4.59+97
Set those in llama.cpp, then:
llm-benchmark benchmark --model "unsloth/Qwen3.5-9B-MTP-GGUF" --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.