Benchmark result
Ministral-3-8B-Instruct-2512-Q4_K_M on NVIDIA GeForce RTX 4060 Laptop GPU — 35.4 tok/s
Measured with llama.cpp b11985-6d194531e on October 31, 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
1st fastest of 6 runs of this model on NVIDIA GeForce RTX 4060 Laptop GPU · median 35.0 tok/s.
sort
- 135.4tok/s63.1llama.cpp · Q4_K_Mthis run
- 235.4tok/s66.2llama.cpp · Q4_K_M
- 335.3tok/s62.4llama.cpp · Q4_K_M
- 434.8tok/s63.6llama.cpp · Q4_K_M
- 534.3tok/s70.1llama.cpp · Q4_K_M
- 634.1tok/s60.8llama.cpp · Q4_K_M
Reproduce this run
toolllama.cpp b11985-6d194531e
modelMinistral-3-8B-Instruct-2512-Q4_K_M (Q4_K_M)
context32,768
think limitnone
samplingtemperature 0.800000011920929 · top_p 0.949999988079071 · top_k 40 · min_p 0.05000000074505806 · presence_penalty 0 · repeat_penalty 1
clientv0.4.98+102
Set those in llama.cpp, then:
llm-benchmark benchmark --model "Ministral-3-8B-Instruct-2512-Q4_K_M" --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.