Benchmark result
Qwen3.6-35B-A3B-IQ4_NL on AMD Radeon RX 7900 XTX — 172.1 tok/s
Measured with llama.cpp b182-c1d0e7a on August 30, 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
4th fastest of 17 runs of this model on AMD Radeon RX 7900 XTX · median 163.2 tok/s.
sort
- 1174.2tok/s79.3llama.cpp
- 2173.9tok/s71.1llama.cpp
- 3173.4tok/s69.9llama.cpp
- 4172.1tok/s79.1llama.cppthis run
- 5169.5tok/s73.4llama.cpp
- 6165.0tok/s73.6llama.cpp
- 7163.6tok/s73.4llama.cpp
- 8163.5tok/s78.1llama.cpp
- 9163.2tok/s72.8llama.cpp
- 10162.6tok/s77.1llama.cpp
- 11161.9tok/s76.7llama.cpp
- 12161.2tok/s83.5llama.cpp
- 13160.9tok/s72.5llama.cpp
- 14160.9tok/s76.9llama.cpp
- 15160.5tok/s74.2llama.cpp
- 16159.9tok/s76.7llama.cpp
- 17154.9tok/s81.8llama.cpp
Reproduce this run
toolllama.cpp b182-c1d0e7a
modelQwen3.6-35B-A3B-IQ4_NL
context65,536
thinkingon
clientv0.4.44+97
Set those in llama.cpp, then:
llm-benchmark benchmark --model "Qwen3.6-35B-A3B-IQ4_NL" --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.