Machine
Gtx 1050 Ti + RX 7800 Xt · Core i7-7700K #c6793251
76 runs of 26 models since Oct 4, 2026. Last run Nov 18, 2026.
76
runs
26
models
59
configs
11
configs run 2+ times
Hardware
- GPU
- NVIDIA GeForce GTX 1050 Ti + AMD Radeon RX 7800 XT 20 GB
- CPU
- Intel(R) Core(TM) i7-7700K CPU @ 4.20GHz 4 cores
- RAM
- 31 GB
- OS
- CachyOS
- Tools
- llama.cpp 44 · llama.cpp k8 v4 -1 budget 9 · OpenAI-compatible 4 · llama.cpp k8 v4 4 · llama.cpp k8 vTurbo4 1 · llama.cpp k4 v4 1 · llama.cpp k5 v5 16384 budget 1 · llama.cpp k8 v4 16384 budget 1 · llama.cpp k8 v4 12288 budget 1 · llama.cpp k4 v4 8192 budget 1 · llama.cpp k4 v4 4096 budget 1 · llama.cpp k4 v4 2048 budget 1 · llama.cpp k5 v5 1 · llama.cpp k8 v8 -1 budget 1 · llama.cpp q5 q5 -1 budget 1 · llama.cpp q8 q4 -1 budget 1 · llama.cpp q8 q8 -1 budget 1 · llama.cpp q8 q8 1 · Strata 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, 26 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
byteshape/Qwen3.8-27B-GGUF
8 runs · 6 configs
★ bestIQ4_XS-ASCII-Condensedllama.cpp64k ctxthinking oneffort xhigh
×2 runs
89.789.0–90.4
steady
41.6 tok/s
41.3–41.9
unsloth/Qwen3.8-27B-UD-Q2_K_XL
6 runs · 2 configs
★ bestQ2_Kllama.cpp64k ctxthinking oneffort xhigh
×5 runs
89.085.9–90.4
steady
45.4 tok/s
44.2–46.0
Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtp-ASCII-Condensed
1 run · 1 config
Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtp
3 runs · 3 configs
Qwen3.8-27B-GSQ-RCO-BuffedMod-IQ3_S_XL-mtp-Q6K
2 runs · 2 configs
Swift-1.5-IQ3_XSS
1 run · 1 config
Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp
14 runs · 12 configs
★ bestllama.cpp128k ctxthinking defaulteffort xhigh
×3 runs
86.685.6–87.3
steady
51.5 tok/s
47.4–52.2
Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensed
7 runs · 6 configs
★ bestllama.cpp k8 v4 -1 budget125k ctxthinking on
×2 runs
86.683.4–89.8
steady
49.4 tok/s
49.0–49.8
Qwen3.8-27B-GSQ-RCO-IQ3_S-recovered
4 runs · 2 configs
★ bestllama.cpp64k ctxthinking defaulteffort xhigh
×3 runs
86.283.8–86.7
steady
34.1 tok/s
33.8–40.7
Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp
5 runs · 4 configs
★ bestllama.cpp64k ctxthinking defaulteffort xhigh
×2 runs
85.683.1–88.1
steady
36.2 tok/s
34.6–37.7
Compare with another machine
Run history
Nov 18, 2026
17:37Qwen3.8-27B-GGUF:IQ4_XS-ASCII-Condensedllama.cpp41.3 tok/s89.005:40Qwen3.8-27B-GGUF:IQ4_XS-ASCII-Condensedllama.cpp42.9 tok/s87.505:08Qwen3.8-27B-GGUF:IQ4_XS-ASCII-Condensedllama.cpp41.9 tok/s90.403:20Qwen3.8-27B-GGUF:IQ4_XSllama.cpp45.6 tok/s89.402:36SenseNova-Loop-9B-Q4_K_Mllama.cpp28.3 tok/s75.5Nov 16, 2026
22:02Qwen3.8-27B-UD-Q2_K_XL×5llama.cpp45.4 tok/s89.019:12Qwen3.8-27B-UD-Q2_K_XLllama.cpp50.6 tok/s73.718:52Qwen3.8-27B-GGUF:IQ4_XSllama.cpp44.3 tok/s84.6Nov 7, 2026
17:13Qwen3.8-27B-GGUF:IQ3_XSllama.cpp45.8 tok/s84.815:28Qwen3.8-27B-GGUF:IQ3_XSllama.cpp45.3 tok/s82.6Show all 76 runs ▾Show fewer ▴
Oct 19, 2026
17:13OrcaCyber-Nano-1.0-AI-Defense-Mix-13Gllama.cpp k8 vTurbo444.4 tok/s83.712:01OrcaCyber-Nano-1.0-AI-Defense-Mix-13Gllama.cpp k4 v442.7 tok/s81.6Oct 18, 2026
22:05MiMo-V2.6-Distill-Qwen-9B-Q8_0OpenAI-compatible51.8 tok/s62.421:41Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_S-mtpllama.cpp k5 v5 16384 budget39.6 tok/s87.421:17Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp k8 v4 16384 budget45.0 tok/s84.020:57Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp k8 v4 12288 budget47.4 tok/s87.420:41Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp k4 v4 8192 budget49.4 tok/s85.420:14Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp k4 v4 4096 budget47.7 tok/s86.320:03Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp k4 v4 2048 budget47.8 tok/s80.001:45Qwen3.8-27B-GSQ-RCO-BuffedMod-IQ3_S_XL-mtp-Q6Kllama.cpp k8 v437.9 tok/s87.801:04Qwen3.8-27B-GSQ-RCO-BuffedMod-IQ3_S_XL-mtp-Q6Kllama.cpp k8 v442.7 tok/s86.8Oct 17, 2026
18:37Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensedllama.cpp k8 v438.4 tok/s84.518:16Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensedllama.cpp k5 v537.4 tok/s87.3Oct 13, 2026
17:47Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtp-ASCII-Condensedllama.cpp k8 v4 -1 budget35.9 tok/s88.917:08Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_S-mtpllama.cpp k8 v4 -1 budget33.7 tok/s86.1Oct 12, 2026
16:34Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_S-mtpllama.cpp k8 v4 -1 budget33.6 tok/s82.615:49Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtpllama.cpp k8 v4 -1 budget31.5 tok/s86.914:46Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtpllama.cpp k8 v4 -1 budget46.4 tok/s82.914:23Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtpOpenAI-compatible34.6 tok/s88.8Oct 9, 2026
10:35Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp-ASCII-Condensedllama.cpp k8 v440.4 tok/s85.210:05Ornith-1.5-35B-A3B.APEX-I-NanoPlus-ASCII-Condensed×3llama.cpp111 tok/s82.009:41Ornith-1.5-35B-A3B-GSQ-RCO-3bit-ASCII-Condensedllama.cpp105 tok/s73.709:36Ornith-1.5-35B-A3B-GSQ-RCO-3bit-ASCII-Condensedllama.cpp k8 v4 -1 budget107 tok/s64.609:17Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensedllama.cpp k8 v4 -1 budget35.4 tok/s83.708:26Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensedllama.cpp k8 v8 -1 budget38.7 tok/s85.507:48Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensed×2llama.cpp k8 v4 -1 budget49.4 tok/s86.605:57Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensedllama.cpp q5 q5 -1 budget50.2 tok/s84.405:22Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp-ASCII-Condensedllama.cpp q8 q4 -1 budget50.6 tok/s88.804:41Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp q8 q8 -1 budget50.6 tok/s86.6Oct 8, 2026
22:45Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp q8 q849.5 tok/s80.819:26qwen3.8-flash-next-coder-iq1_mStrata42.2 tok/s63.8Oct 6, 2026
10:11Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ2_S-mtpllama.cpp36.1 tok/s78.709:39Qwen3.8-27B-GSQ-RCO-IQ3_S-recoveredllama.cpp33.8 tok/s83.808:52Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_S-mtpllama.cpp34.6 tok/s83.107:23Darwin-28B-REASON.i1-IQ3_Sllama.cpp21.1 tok/s71.006:49Darwin-28B-REASON.i1-IQ3_XXSllama.cpp38.4 tok/s71.206:44Darwin-28B-REASON.i1-IQ3_XXSllama.cpp38.4 tok/s73.1Oct 5, 2026
17:12Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp41.0 tok/s86.416:05Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp×2llama.cpp51.8 tok/s87.014:57Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_S-mtpllama.cpp37.2 tok/s79.014:27Nex-N2.5-mini-IQ3_XXSllama.cpp92.3 tok/s80.6Oct 4, 2026
23:46Nex-N2.5-mini-IQ3_XXSllama.cpp116 tok/s84.323:28Qwen3.8-27B-Opus-Distill-IQ3_XXS×2llama.cpp28.9 tok/s76.323:04Qwen3.8-27B-GSQ-RCO-IQ3_S-recoveredllama.cpp33.3 tok/s80.122:05Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp48.5 tok/s87.620:40Qwen3.8-27B-GSQ-RCO-IQ3_S-recovered×2llama.cpp37.4 tok/s86.418:52Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp52.8 tok/s77.718:31Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp48.3 tok/s82.718:07Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtpllama.cpp47.4 tok/s85.617:45Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_S-mtpllama.cpp48.9 tok/s79.717:17Swift-1.5-Qwen3.8-27B-GSQ-RCO-IQ3_S-mtpllama.cpp37.7 tok/s88.116:08Nex-N2.5-mini-IQ3_XXSllama.cpp98.9 tok/s84.115:57Ornith-1.5-35B-A3B-GSQ-RCO-3bitllama.cpp91.5 tok/s82.1