Benchmark result
swift-1.5-9b on AMD Radeon RX 9050 / 9060 XT — 32.2 tok/s
Measured with llama.cpp manual on November 12, 2026.
Support llm-bench.io
llm-bench.io is free. If it helped you, I'd appreciate your support to cover the infrastructure costs.
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 AMD Radeon RX 9050 / 9060 XT · median 41.7 tok/s.
sort
- 148.6tok/s75.5llama.cpp
- 248.5tok/s57.0llama.cpp
- 348.5tok/s69.6llama.cpp
- 443.8tok/s56.4llama.cpp
- 543.7tok/s75.8llama.cpp
- 643.7tok/s70.9llama.cpp
- 743.7tok/s75.4llama.cpp
- 839.7tok/s68.7llama.cpp
- 939.7tok/s70.0llama.cpp
- 1039.7tok/s75.0llama.cpp
- 1139.6tok/s73.4llama.cpp
- 1232.2tok/s77.9llama.cpp
- 1332.2tok/s78.4llama.cpp
- 1432.2tok/s77.8llama.cppthis run
Reproduce this run
toolllama.cpp manual
modelswift-1.5-9b
context200,000
thinkingon
think limit12,288 tokens
samplingtemperature 1 · top_p 0.949999988079071 · top_k 20 · min_p 0 · presence_penalty 1.5 · repeat_penalty 1
clientv0.4.98+102
Set those in llama.cpp, then:
llm-benchmark benchmark --model "swift-1.5-9b" --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.