Machine
2× Tesla V100 Pcie 32gb · Xeon W-2133 #48decc36
117 runs of 47 models since Nov 1, 2026. Last run Nov 10, 2026.
117
runs
47
models
63
configs
34
configs run 2+ times
Hardware
- GPU
- 2× Tesla V100-PCIE-32GB 64 GB
- CPU
- Intel(R) Xeon(R) W-2133 CPU @ 3.60GHz 6 cores
- RAM
- 46 GB
- OS
- Arch Linux
- Tools
- llama.cpp 114 · OpenAI-compatible 2 · vLLM 1
- Client
- 0.4.99
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, 47 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
Qwen3.8-27B-GSQ-RCO
2 runs · 1 config
★ bestllama.cpp32k ctxthinking defaulteffort xhigh
×2 runs
87.686.2–88.9
steady
41.1 tok/s
39.0–43.1
Arete-27B_t1p95k20m0
2 runs · 1 config
★ bestllama.cpp32k ctxthinking defaulteffort xhigh
×2 runs
86.586.4–86.6
steady
37.2 tok/s
37.2–37.2
Cyber-Tiel_t06p95k20m0
2 runs · 1 config
★ bestllama.cpp32k ctxthinking default
×2 runs
86.184.9–87.2
steady
106 tok/s
105–107
Sabi-Noema_t1p95k20m0
2 runs · 1 config
★ bestllama.cpp32k ctxthinking defaulteffort xhigh
×2 runs
85.985.7–86.2
steady
58.9 tok/s
58.3–59.4
Qwen3.8-27B-GSQ-RCO_t1p95k20m0
2 runs · 1 config
★ bestllama.cpp32k ctxthinking defaulteffort xhigh
×2 runs
85.884.1–87.6
steady
41.4 tok/s
41.1–41.6
Swift-1.5-Uncensored
6 runs · 1 config
★ bestllama.cpp32k ctxthinking defaulteffort xhigh
×6 runs
85.680.7–88.3
steady
60.8 tok/s
42.7–69.1
Cyber-Tiel
5 runs · 2 configs
★ bestllama.cpp32k ctxthinking default
×4 runs
85.183.0–86.4
steady
107 tok/s
106–109
Swift-1.5-Uncensored_t1p95k20m0
2 runs · 1 config
★ bestllama.cpp32k ctxthinking defaulteffort xhigh
×2 runs
85.183.8–86.5
steady
60.1 tok/s
59.8–60.4
GRM-3.2-Sky
4 runs · 1 config
★ bestllama.cpp32k ctxthinking default
×4 runs
84.683.1–86.2
steady
99.0 tok/s
94.4–99.5
Thinklite-27-GGUF_t1p95k20m0
3 runs · 2 configs
★ bestllama.cpp32k ctxthinking defaulteffort xhigh
×2 runs
84.680.6–88.6
steady
48.4 tok/s
48.3–48.5
Compare with another machine
Run history
Nov 10, 2026
20:09Qwen-MidTempllama.cpp229 tok/s32.020:07Qwen-MidTempllama.cpp238 tok/s19.520:06Qwen-LowTempllama.cpp219 tok/s21.420:03Qwen-LowTempllama.cpp232 tok/s28.919:35Thinklite-27-GGUF_t1p95k20m0×2llama.cpp48.4 tok/s84.619:12Terminal-RSI_t06p95k20m0×2llama.cpp66.7 tok/s76.219:04Swift-1.5-Uncensored_t1p95k20m0×2llama.cpp60.1 tok/s85.118:37Sage-A1.5_t1p95k20m0P15×2llama.cpp93.0 tok/s75.918:19Sabi-Noema_t1p95k20m0×2llama.cpp58.9 tok/s85.917:57Qwen3.8-27B-GSQ-RCO_t1p95k20m0×2llama.cpp41.4 tok/s85.817:21Qwen3-Coder-Next_t1p95k40m0×2llama.cpp81.5 tok/s78.517:11Qwen-Turbo-Uncensored_t1p95k20m0×2llama.cpp39.7 tok/s70.716:52Qwen-EfficientThink_t1p95k20m0×2llama.cpp38.5 tok/s82.316:25Nemotron-3.5-Lightning_t1p95k40m05×2llama.cpp126 tok/s75.616:14Nanbeige4.3_t1p95k0m0×2llama.cpp30.9 tok/s72.515:28Muse-Glimmer-30B_t1p95k64m05×2llama.cpp40.6 tok/s84.515:14Muse-Glimmer-30B-Heretic_t1p95k64m05×2llama.cpp52.3 tok/s78.415:01Kwaipilot-KAT-Coder_t1p95k20m0×2llama.cpp99.9 tok/s80.914:49Gemma4-31B-IT_t1p95k64m05×2llama.cpp51.4 tok/s75.114:40GRM-3.2-Sky_t1p95k20m0×2llama.cpp99.8 tok/s78.414:31GLM-4.7-Flash-REAP_t02p95k50m01D11×2llama.cpp49.7 tok/s65.614:18Flint_t1p95k20m0×2llama.cpp88.9 tok/s75.514:04Cyber-Tiel_t06p95k20m0×2llama.cpp106 tok/s86.113:54Arete-27B_t1p95k20m0×2llama.cpp37.2 tok/s86.513:13Apex-35B-V1.0_t1p95k20m0×2llama.cpp99.3 tok/s56.612:49Qwen-HighTempllama.cpp237 tok/s20.112:44Qwen-LowTemp×2llama.cpp213 tok/s26.012:34Qwen-HighTemp×4llama.cpp235 tok/s25.712:06Qwen-LowTemp×2llama.cpp232 tok/s29.012:00Qwen-LowTemp×2llama.cpp222 tok/s19.311:54Qwen-LowTemp×2llama.cpp230 tok/s23.911:52Qwen-HighTempllama.cpp235 tok/s24.311:50Qwen-HighTemp×3llama.cpp236 tok/s26.311:10Thinklite-27-GGUF_t1p95k20m0llama.cpp49.0 tok/s70.410:24Kwaipilot-KAT-Coder_t1p95k20m0×3llama.cpp101 tok/s76.909:03Kwaipilot-KAT-Coder_t1p95k20m0×3llama.cpp101 tok/s78.300:54Apex-35B-V1.0_t1p95k20m0llama.cpp98.8 tok/s57.9Nov 9, 2026
22:05Kwaipilot-KAT-Coder×2OpenAI-compatible81.1 tok/s82.320:05Swift-1.5-Uncensored×3llama.cpp60.3 tok/s85.415:51Kwaipilot-KAT-Coder×3llama.cpp81.2 tok/s66.715:22GRM-3.2-Sky×3llama.cpp99.4 tok/s84.414:53Cyber-Tiel×3llama.cpp107 tok/s84.5Nov 4, 2026
09:30Sage-A1.5llama.cpp132 tok/s80.509:20Swift-1.5-Uncensoredllama.cpp65.4 tok/s80.709:09Flintllama.cpp96.0 tok/s76.4Nov 3, 2026
21:47Swift-1.5-Uncensoredllama.cpp69.1 tok/s88.321:36Muse-Glimmer-30B-Hereticllama.cpp61.4 tok/s83.021:29Muse-Glimmer-30Bllama.cpp54.3 tok/s80.421:18Sabi-Noemallama.cpp65.8 tok/s82.521:10Cyber-Tielllama.cpp107 tok/s86.421:00Sagellama.cpp128 tok/s82.420:53Terminal-RSIllama.cpp70.2 tok/s81.120:33Swift-1.5-Uncensoredllama.cpp42.7 tok/s86.420:02RavenX-Chaos-Agentllama.cpp47.0 tok/s35.219:45Qwen3.8-27B-GSQ-RCOllama.cpp43.1 tok/s88.919:25Qwen3-Coder-Nextllama.cpp77.8 tok/s81.019:20Qwen-Turbo-Uncensoredllama.cpp42.6 tok/s74.219:11Qwen-EfficientThinkllama.cpp40.4 tok/s83.018:52Nemotron-3.5-Lightningllama.cpp126 tok/s80.118:46Nanabeigellama.cpp32.2 tok/s75.818:24Kwaipilot-KAT-Coderllama.cpp97.1 tok/s81.618:19Gemma4-31B-ITllama.cpp58.1 tok/s77.518:07GRM-3.2-Skyllama.cpp94.4 tok/s84.817:56GLM-4.7-Flash-REAPllama.cpp58.4 tok/s66.316:34GLM-4.7-Flash×2llama.cpp64.4 tok/s66.3Nov 2, 2026
07:05qwen3.8-27b-dflash2vLLM112 tok/s69.4