Benchmark result
openbmb/MiniCPM5-2B-GGUF on NVIDIA GeForce RTX 4070 — 206.9 tok/s
Measured with llama.cpp b11541-ce8c04021 on October 14, 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
3rd fastest of 9 runs of this model on NVIDIA GeForce RTX 4070 · median 174.8 tok/s.
sort
- 1235.0tok/s31.5Unsloth Studio · Q4_K_M · KV f16
- 2224.7tok/s46.6llama.cpp
- 3206.9tok/s47.3llama.cppthis run
- 4185.8tok/s51.2llama.cpp
- 5174.8tok/s44.4Unsloth Studio · Q4_K_M · KV f16
- 6167.8tok/s47.9Unsloth Studio · Q4_K_M · KV q8_0
- 7166.1tok/s47.4Unsloth Studio · Q6_K_L · KV f16
- 8154.5tok/s50.7Unsloth Studio · Q4_K_M · KV f16
- 9129.1tok/s40.4Unsloth Studio · Q6_K_L · KV q8_0
openbmb/MiniCPM5-2B-GGUF on other hardware
Reproduce this run
toolllama.cpp b11541-ce8c04021
modelopenbmb/MiniCPM5-2B-GGUF
context131,072
thinkingon
think limitnone
samplingtemperature 1 · top_p 0.949999988079071 · top_k 40 · min_p 0.05000000074505806 · presence_penalty 0 · repeat_penalty 1
clientv0.4.93+102
Set those in llama.cpp, then:
llm-benchmark benchmark --model "openbmb/MiniCPM5-2B-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.