Benchmark result
Cyber-Tiel-Coder-35B-A3B-MTP-UD-Q4_K_XL on Tesla P100-PCIE-16GB — 36.6 tok/s
Measured with llama.cpp b1-5266f24 on September 22, 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
14th fastest of 14 runs of this model on Tesla P100-PCIE-16GB · median 38.6 tok/s.
sort
- 142.1tok/s85.8llama.cpp · Q4_K
- 241.4tok/s84.6llama.cpp · Q4_K
- 341.3tok/s82.1llama.cpp · Q4_K
- 440.4tok/s73.7llama.cpp · Q4_K
- 539.9tok/s81.0llama.cpp · Q4_K
- 639.6tok/s77.6llama.cpp · Q4_K
- 738.7tok/s81.4llama.cpp · Q4_K
- 838.4tok/s80.1llama.cpp · Q4_K
- 938.3tok/s82.4llama.cpp · Q4_K
- 1038.0tok/s85.6llama.cpp · Q4_K
- 1137.8tok/s82.9llama.cpp · Q4_K
- 1237.7tok/s84.9llama.cpp · Q4_K
- 1337.6tok/s77.8llama.cpp · Q4_K
- 1436.6tok/s82.2llama.cpp · Q4_Kthis run
Reproduce this run
toolllama.cpp b1-5266f24
modelCyber-Tiel-Coder-35B-A3B-MTP-UD-Q4_K_XL (Q4_K)
context65,536
thinkingon
samplingtemperature 0.6000000238418579 · top_p 0.949999988079071 · top_k 20 · min_p 0 · presence_penalty 0 · repeat_penalty 1
clientv0.4.71+98
Set those in llama.cpp, then:
llm-benchmark benchmark --model "Cyber-Tiel-Coder-35B-A3B-MTP-UD-Q4_K_XL" --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.