Benchmark result

JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:MXFP4_MOE on AMD Radeon RX 9070/9070 XT/9070 GRE — 161.9 tok/s

Measured with llama.cpp b11388-050439614 on October 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

5th fastest of 9 runs of this model on AMD Radeon RX 9070/9070 XT/9070 GRE · median 161.9 tok/s.

sort

Compare the nearest 5 →

Reproduce this run

toolllama.cpp b11388-050439614

modelJetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:MXFP4_MOE

context32,768

think limit8,192 tokens

samplingtemperature 0.4000000059604645 · top_p 0.949999988079071 · top_k 20 · min_p 0.10000000149011612 · presence_penalty 1.5 · repeat_penalty 1

clientv0.4.89+102

Set those in llama.cpp, then:

llm-benchmark benchmark --model "JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF:MXFP4_MOE" --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.