Benchmark result
KAT-Coder-V2.5-Dev.i1-IQ3_XXS on 2× NVIDIA GeForce RTX 5060 — 86.9 tok/s
Measured with llama.cpp b11217-c9064dded on September 28, 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.
How this run compares
1st fastest of 2 runs of this model on 2× NVIDIA GeForce RTX 5060 · Multi-GPU · median 85.7 tok/s.
sort
- 186.9tok/s77.1llama.cppthis run
- 284.6tok/s76.5llama.cpp
Reproduce this run
toolllama.cpp b11217-c9064dded
modelKAT-Coder-V2.5-Dev.i1-IQ3_XXS
context131,072
thinkingon
samplingtemperature 0.6000000238418579 · top_p 0.8500000238418579 · top_k 20 · min_p 0 · presence_penalty 0 · repeat_penalty 1.0499999523162842
clientv0.4.82+98
Set those in llama.cpp, then:
llm-benchmark benchmark --model "KAT-Coder-V2.5-Dev.i1-IQ3_XXS" --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.