Benchmark result
Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp on AMD Radeon RX 7800 XT — 45.0 tok/s
Measured with llama.cpp k8 v4 16384 budget manual on October 18, 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 15 runs of this model on AMD Radeon RX 7800 XT · median 48.3 tok/s.
sort
- 152.8tok/s77.7llama.cpp
- 252.2tok/s86.6llama.cpp
- 351.5tok/s87.3llama.cpp
- 450.6tok/s86.6llama.cpp q8 q8 -1 budget
- 549.5tok/s80.8llama.cpp q8 q8
- 649.4tok/s85.4llama.cpp k4 v4 8192 budget
- 748.5tok/s87.6llama.cpp
- 848.3tok/s82.7llama.cpp
- 947.8tok/s80.0llama.cpp k4 v4 2048 budget
- 1047.7tok/s86.3llama.cpp k4 v4 4096 budget
- 1147.4tok/s87.4llama.cpp k8 v4 12288 budget
- 1247.4tok/s85.6llama.cpp
- 1346.6tok/s87.1llama.cpp
- 1445.0tok/s84.0llama.cpp k8 v4 16384 budgetthis run
- 1541.0tok/s86.4llama.cpp
Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp on other hardware
Reproduce this run
toolllama.cpp k8 v4 16384 budget manual
modelSwift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp
context128,000
thinkingon
clientv0.4.92+102
Set those in llama.cpp k8 v4 16384 budget, then:
llm-benchmark benchmark --model "Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp" --tool "llama.cpp k8 v4 16384 budget"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.