Machine
Ai Pro R9700 · Ryzen 9 5950X #9e7a32e4
44 runs of 14 models since Oct 1, 2026. Last run Nov 7, 2026.
44
runs
14
models
40
configs
3
configs run 2+ times
Hardware
- GPU
- AMD Radeon AI PRO R9700 32 GB
- CPU
- AMD Ryzen 9 5950X 16-Core Processor 16 cores
- RAM
- 60 GB
- OS
- Ubuntu 26.04.1 LTS
- Tools
- llama.cpp 43 · Strata 1
- Client
- 0.4.98
Models on this machine
One dot per model, at its best config. Up is better answers, right is faster. Time to finish counts every token the model wrote, thinking included, so a model that talks a lot ends up further left.
Best trade-off: quality 60+, and nothing else here is both faster and betterOther models · dot size = number of runs
Medians of each model's best config, 14 models.
Best config per model
sort
Runs with the same model, quant, tool, context, thinking and KV cache count as one config. Numbers are the median of its runs, and each dot is one run on the 0–100 quality scale. ★ marks the best config with at least 2 rated runs.
configrunsqualityeach runspeed
unsloth/Muse-Glimmer-30B-GGUF
6 runs · 6 configs
unsloth/Qwen3.8-27B-GGUF
6 runs · 5 configs
★ bestUD-Q4_K_XLllama.cpp64k ctxthinking oneffort medium
×2 runs
84.483.7–85.1
steady
35.8 tok/s
30.0–41.5
bartowski/NousResearch_Arete-27B-GGUF
1 run · 1 config
Qwen3.6-27B-Q4_K_M
1 run · 1 config
ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF
1 run · 1 config
unsloth/gemma-4-31B-it-GGUF
2 runs · 2 configs
unsloth/Qwen3.6-35B-A3B-MTP-GGUF
5 runs · 4 configs
★ bestUD-Q4_K_XLllama.cpp64k ctxthinking on
×2 runs
77.172.5–81.7
varies
101 tok/s
101–101
JetBrains/Mellum2.1-12B-A2.5B-Thinking-GGUF
3 runs · 3 configs
unsloth/gemma-4-26B-A4B-it-GGUF
6 runs · 6 configs
ibm-granite/granite-4.2-30b-GGUF
1 run · 1 config
Compare with another machine
Run history
Oct 17, 2026
11:58GLM-4.7-Flash-GGUF:UD-Q4_K_XLllama.cpp118 tok/s67.311:19Mellum2.1-12B-A2.5B-Thinking-GGUF:Q8_0llama.cpp180 tok/s68.611:16Mellum2.1-12B-A2.5B-Thinking-GGUF:Q8_0llama.cpp179 tok/s60.411:08Mellum2.1-12B-A2.5B-Thinking-GGUF:Q4_K_Mllama.cpp232 tok/s76.0Oct 13, 2026
21:27gemma-4-26B-A4B-it-GGUF:UD-Q5_K_XLllama.cpp69.8 tok/s74.221:24gemma-4-26B-A4B-it-GGUF:UD-Q5_K_XLllama.cpp69.5 tok/s75.821:12gemma-4-31B-it-qat-GGUF:UD-Q4_K_XLllama.cpp53.7 tok/s73.521:10gemma-4-31B-it-qat-GGUF:UD-Q4_K_XLllama.cpp54.0 tok/s69.621:07gemma-4-31B-it-qat-GGUF:UD-Q4_K_XLllama.cpp53.7 tok/s70.821:04gemma-4-31B-it-qat-GGUF:UD-Q4_K_XLllama.cpp53.7 tok/s78.520:58gemma-4-31B-it-qat-GGUF:UD-Q4_K_XLllama.cpp53.4 tok/s67.020:53Muse-Glimmer-30B-GGUF:UD-Q4_K_XLllama.cpp34.4 tok/s83.220:43Muse-Glimmer-30B-GGUF:UD-Q4_K_XLllama.cpp34.5 tok/s85.820:06Muse-Glimmer-30B-GGUF:UD-Q4_K_XLllama.cpp34.6 tok/s84.6Oct 11, 2026
12:49gemma-4-31B-it-GGUF:UD-Q4_K_XLllama.cpp30.5 tok/s77.212:39gemma-4-31B-it-GGUF:UD-Q4_K_XLllama.cpp31.7 tok/s71.112:19gemma-4-31B-it-qat-GGUF:UD-Q4_K_XL×2llama.cpp37.1 tok/s69.911:21starcoder2-15b-instruct-v0.1-GGUFllama.cpp54.7 tok/s19.9Show all 44 runs ▾Show fewer ▴
Oct 10, 2026
21:54Qwen3.8-27B-GGUF:UD-Q4_K_XLllama.cpp41.5 tok/s83.721:44Qwen3.8-27B-GGUF:UD-Q4_K_XLllama.cpp41.0 tok/s69.3Oct 4, 2026
17:03Qwen3.8-27B-GGUF:UD-Q4_K_XLllama.cpp29.4 tok/s75.416:21Muse-Glimmer-30B-GGUF:UD-Q4_K_XLllama.cpp33.8 tok/s79.516:02gemma-4-26B-A4B-it-GGUF:UD-Q5_K_XLllama.cpp64.2 tok/s63.615:58gemma-4-26B-A4B-it-GGUF:UD-Q5_K_XLllama.cpp64.7 tok/s67.415:33gemma-4-26B-A4B-it-GGUF:UD-Q4_K_XLllama.cpp64.6 tok/s68.915:26gemma-4-26B-A4B-it-GGUF:UD-Q4_K_XLllama.cpp65.1 tok/s75.715:14Qwen3.6-35B-A3B-MTP-GGUF:UD-Q5_K_XLllama.cpp98.3 tok/s75.815:08Qwen3.6-35B-A3B-MTP-GGUF:UD-Q5_K_XLllama.cpp99.4 tok/s81.114:35Qwen3.6-35B-A3B-MTP-GGUF:UD-Q4_K_XL×2llama.cpp101 tok/s77.114:19Qwen3.6-35B-A3B-MTP-GGUF:UD-Q4_K_XLllama.cpp101 tok/s84.313:59GLM-4.7-Flash-GGUF:UD-Q4_K_XLllama.cpp112 tok/s62.413:55GLM-4.7-Flash-GGUF:UD-Q4_K_XLllama.cpp111 tok/s66.1