#1 · Best device result · mlx · BASELINE
TaterTotterson/Gemma-4-26B-A4B-IT-UD-Q4_K_XL-mlx-Tater-NoThink 674e8a504329b470c761c2e1feeb31d93d6ceeb7
Text only Declared model inputs · not scored
88.6Tater score
Tater Ready All Tater readiness gates pass.
88.6 raw formula96.1% required accuracy55.09 tok/s0.23s TTFT18.5 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware1 tested device1 unique run2 observations
Chat 97.14%Routing 95.42%Spudex (optional) 97.58%Synthesis 90.00%Tool Accuracy 100.00%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:41:01.118063+00:00 2 Tater Ready 88.6 88.6 96.1% 55.09 0.23s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 86.5 accuracy + 1.5 speed + 0.6 TTFT + 0.1 memory + 0.0 readiness gate
#2 · Best device result · llama.cpp · MTP
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
86.8Tater score
Tater Ready All Tater readiness gates pass.
86.8 raw formula93.0% required accuracy104.57 tok/s0.54s TTFT18.9 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware3 tested devices1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:29:18.352445+00:00 2 Tater Ready 86.8 86.8 93.0% 104.57 0.54s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 2.8 speed + 0.2 TTFT + 0.1 memory + 0.0 readiness gate
#3 · Best device result · llama.cpp · BASELINE
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
86.1Tater score
Tater Ready All Tater readiness gates pass.
86.1 raw formula93.0% required accuracy79.73 tok/s0.53s TTFT18.2 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware3 tested devices1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:28:50.368818+00:00 2 Tater Ready 86.1 86.1 93.0% 79.73 0.53s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 2.1 speed + 0.2 TTFT + 0.1 memory + 0.0 readiness gate
#4 · Best device result · llama.cpp · DFLASH
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
85.3Tater score
Tater Ready All Tater readiness gates pass.
85.3 raw formula93.0% required accuracy48.10 tok/s0.74s TTFT4.4 GiB peak RSSAMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:38:12.398047+00:00 2 Tater Ready 85.3 85.3 93.0% 48.10 0.74s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 1.3 speed + 0.2 TTFT + 0.2 memory + 0.0 readiness gate
#5 · Best device result · llama.cpp · BASELINE
unsloth/gemma-4-12B-it-qat-GGUF gemma-4-12B-it-qat-UD-Q4_K_XL.gguf
Vision Video Audio Declared model inputs · not scored
84.4Tater score
Tater Ready All Tater readiness gates pass.
84.4 raw formula91.7% required accuracy62.96 tok/s1.22s TTFT8.5 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 100.00%Routing 90.08%Spudex (optional) 77.86%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:39:32.847663+00:00 2 Tater Ready 84.4 84.4 91.7% 62.96 1.22s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 82.5 accuracy + 1.7 speed + 0.1 TTFT + 0.1 memory + 0.0 readiness gate
#6 · Best device result · llama.cpp · BASELINE
TaterTotterson/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF-Tater-NoThink NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Tater-NoThink-Q4_0.gguf
Vision Video Declared model inputs · not scored
84.4Tater score
Tater Ready All Tater readiness gates pass.
84.4 raw formula90.6% required accuracy97.59 tok/s0.66s TTFT16.5 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 97.14%Routing 93.00%Spudex (optional) 82.09%Synthesis 90.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:30:52.427749+00:00 2 Tater Ready 84.4 84.4 90.6% 97.59 0.66s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 81.6 accuracy + 2.6 speed + 0.2 TTFT + 0.1 memory + 0.0 readiness gate
#7 · Best device result · llama.cpp · MTP
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.4Tater score
Tater Ready All Tater readiness gates pass.
82.4 raw formula88.0% required accuracy111.59 tok/s0.61s TTFT23.1 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware3 tested devices1 unique run2 observations
Chat 100.00%Routing 83.00%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:32:27.413758+00:00 2 Tater Ready 82.4 82.4 88.0% 111.59 0.61s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 79.2 accuracy + 3.0 speed + 0.2 TTFT + 0.0 memory + 0.0 readiness gate
#8 · Best device result · llama.cpp · MTP
TaterTotterson/Qwen3.8-27B-GGUF-Tater-NoThink Qwen3.8-27B-Tater-NoThink-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.2Tater score
Tater Ready All Tater readiness gates pass.
82.2 raw formula90.4% required accuracy21.33 tok/s2.76s TTFT3.8 GiB peak RSSAMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 100.00%Routing 88.42%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:49:43.588718+00:00 2 Tater Ready 82.2 82.2 90.4% 21.33 2.76s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 81.4 accuracy + 0.6 speed + 0.1 TTFT + 0.2 memory + 0.0 readiness gate
#9 · Best device result · llama.cpp · BASELINE
TaterTotterson/Qwen3.8-27B-GGUF-Tater-NoThink Qwen3.8-27B-Tater-NoThink-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.2Tater score
Tater Ready All Tater readiness gates pass.
82.2 raw formula90.4% required accuracy27.18 tok/s3.30s TTFT17.5 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 100.00%Routing 88.42%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:35:34.657738+00:00 2 Tater Ready 82.2 82.2 90.4% 27.18 3.30s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 81.4 accuracy + 0.7 speed + 0.0 TTFT + 0.1 memory + 0.0 readiness gate
#10 · Best device result · llama.cpp · BASELINE
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
81.5Tater score
Tater Ready All Tater readiness gates pass.
81.5 raw formula88.0% required accuracy75.91 tok/s0.61s TTFT20.4 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware3 tested devices1 unique run2 observations
Chat 100.00%Routing 83.00%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:31:46.399367+00:00 2 Tater Ready 81.5 81.5 88.0% 75.91 0.61s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 79.2 accuracy + 2.0 speed + 0.2 TTFT + 0.0 memory + 0.0 readiness gate
#11 · Best device result · llama.cpp · BASELINE
unsloth/gemma-4-E4B-it-qat-GGUF gemma-4-E4B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
79.9Tater score
Limited — Not Ready Routing 75.4% is below 80%.
80.3 raw formula85.2% required accuracy99.13 tok/s0.16s TTFT5.4 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 100.00%Routing 75.42%Spudex (optional) 87.31%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:40:17.672114+00:00 2 Limited — Not Ready 79.9 80.3 85.2% 99.13 0.16s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 76.7 accuracy + 2.6 speed + 0.8 TTFT + 0.2 memory + -0.4 readiness gate
#12 · Best device result · llama.cpp · MTP
unsloth/gemma-4-E2B-it-qat-GGUF gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
79.9Tater score
Limited — Not Ready Required accuracy 84.0% is below 85%. Routing 72.8% is below 80%.
82.3 raw formula84.0% required accuracy181.02 tok/s0.09s TTFT2.2 GiB peak RSSAMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 100.00%Routing 72.79%Spudex (optional) 73.41%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:52:46.830570+00:00 2 Limited — Not Ready 79.9 82.3 84.0% 181.02 0.09s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 75.6 accuracy + 4.8 speed + 1.5 TTFT + 0.4 memory + -2.4 readiness gate
#13 · Best device result · llama.cpp · BASELINE
unsloth/gemma-4-E2B-it-qat-GGUF gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
79.9Tater score
Limited — Not Ready Required accuracy 83.5% is below 85%. Routing 71.7% is below 80%.
80.6 raw formula83.5% required accuracy141.61 tok/s0.09s TTFT3.7 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 100.00%Routing 71.67%Spudex (optional) 73.41%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:39:45.924912+00:00 2 Limited — Not Ready 79.9 80.6 83.5% 141.61 0.09s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 75.2 accuracy + 3.8 speed + 1.4 TTFT + 0.2 memory + -0.7 readiness gate
#14 · Best device result · llama.cpp · BASELINE
unsloth/Qwen3.5-0.8B-GGUF Qwen3.5-0.8B-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
49.9Tater score
Not Fit Required accuracy 67.3% is below 70%. Tool accuracy 62.2% is below 70%. Routing 59.3% is below 70%.
68.6 raw formula67.3% required accuracy229.34 tok/s0.12s TTFT1.2 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 92.14%Routing 59.29%Spudex (optional) 39.51%Synthesis 90.83%Tool Accuracy 62.22%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:40:27.739073+00:00 2 Not Fit 49.9 68.6 67.3% 229.34 0.12s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 60.6 accuracy + 6.1 speed + 1.1 TTFT + 0.8 memory + -18.7 readiness gate
#15 · Best device result · llama.cpp · BASELINE
allenai/OLMoE-1B-7B-0924-Instruct-GGUF olmoe-1b-7b-0924-instruct-q4_k_m.gguf
Vision Video Declared model inputs · not scored
46.3Tater score
Not Fit Required accuracy 41.3% is below 70%. Tool accuracy 44.8% is below 70%. Routing 11.6% is below 70%.
46.3 raw formula41.3% required accuracy262.69 tok/s0.06s TTFT5.8 GiB peak RSSApple M3 Ultra · 96.0 GiB best hardware2 tested devices1 unique run2 observations
Chat 81.43%Routing 11.62%Spudex (optional) 25.38%Synthesis 100.00%Tool Accuracy 44.78%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:28:14.739220+00:00 2 Not Fit 46.3 46.3 41.3% 262.69 0.06s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 37.1 accuracy + 7.0 speed + 2.0 TTFT + 0.1 memory + 0.0 readiness gate
#1 · Device average · llama.cpp · MTP
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
84.5Tater score
Tater Ready All Tater readiness gates pass.
84.5 raw formula93.0% required accuracy26.00 tok/s3.76s TTFT17.2 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD Ryzen AI 9 HX 470 w/ Radeon 890M · 26.5 GiB 2026-08-30T15:39:09.509580+00:00 2 Tater Ready 84.5 84.5 93.0% 26.00 3.76s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 0.7 speed + 0.0 TTFT + 0.1 memory + 0.0 readiness gate
#2 · Device average · llama.cpp · BASELINE
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
84.2Tater score
Tater Ready All Tater readiness gates pass.
84.2 raw formula93.0% required accuracy15.14 tok/s3.80s TTFT16.8 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD Ryzen AI 9 HX 470 w/ Radeon 890M · 26.5 GiB 2026-08-30T15:36:42.881541+00:00 2 Tater Ready 84.2 84.2 93.0% 15.14 3.80s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 0.4 speed + 0.0 TTFT + 0.1 memory + 0.0 readiness gate
#3 · Device average · llama.cpp · BASELINE
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
79.3Tater score
Tater Ready — Provisional All Tater readiness gates pass.
79.3 raw formula88.0% required accuracy2.52 tok/s9.91s TTFT20.5 GiB peak RSS1 device1 unique run1 observation
Chat 100.00%Routing 83.00%Spudex (optional) 82.09%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD Ryzen AI 9 HX 470 w/ Radeon 890M · 26.5 GiB 2026-08-30T15:55:16.045138+00:00 1 Tater Ready — Provisional 79.3 79.3 88.0% 2.52 9.91s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 79.2 accuracy + 0.1 speed + 0.0 TTFT + 0.0 memory + 0.0 readiness gate
#4 · Device average · llama.cpp · MTP
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
79.3Tater score
Tater Ready — Provisional All Tater readiness gates pass.
79.3 raw formula88.0% required accuracy2.28 tok/s24.14s TTFT22.0 GiB peak RSS1 device1 unique run1 observation
Chat 100.00%Routing 83.00%Spudex (optional) 82.09%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD Ryzen AI 9 HX 470 w/ Radeon 890M · 26.5 GiB 2026-08-30T16:21:20.375836+00:00 1 Tater Ready — Provisional 79.3 79.3 88.0% 2.28 24.14s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 79.2 accuracy + 0.1 speed + 0.0 TTFT + 0.0 memory + 0.0 readiness gate
#1 · Device average · llama.cpp · MTP
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
86.2Tater score
Tater Ready All Tater readiness gates pass.
86.2 raw formula93.0% required accuracy75.34 tok/s0.71s TTFT2.5 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:37:10.017856+00:00 2 Tater Ready 86.2 86.2 93.0% 75.34 0.71s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 2.0 speed + 0.2 TTFT + 0.4 memory + 0.0 readiness gate
#2 · Device average · llama.cpp · DFLASH
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
85.3Tater score
Tater Ready All Tater readiness gates pass.
85.3 raw formula93.0% required accuracy48.10 tok/s0.74s TTFT4.4 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:38:12.398047+00:00 2 Tater Ready 85.3 85.3 93.0% 48.10 0.74s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 1.3 speed + 0.2 TTFT + 0.2 memory + 0.0 readiness gate
#3 · Device average · llama.cpp · MTP
TaterTotterson/Qwen3.8-27B-GGUF-Tater-NoThink Qwen3.8-27B-Tater-NoThink-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.2Tater score
Tater Ready All Tater readiness gates pass.
82.2 raw formula90.4% required accuracy21.33 tok/s2.76s TTFT3.8 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 88.42%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:49:43.588718+00:00 2 Tater Ready 82.2 82.2 90.4% 21.33 2.76s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 81.4 accuracy + 0.6 speed + 0.1 TTFT + 0.2 memory + 0.0 readiness gate
#4 · Device average · llama.cpp · MTP
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.0Tater score
Tater Ready All Tater readiness gates pass.
82.0 raw formula88.0% required accuracy77.87 tok/s0.94s TTFT1.6 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 83.00%Spudex (optional) 82.09%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:42:22.436768+00:00 2 Tater Ready 82.0 82.0 88.0% 77.87 0.94s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 79.2 accuracy + 2.1 speed + 0.1 TTFT + 0.6 memory + 0.0 readiness gate
#5 · Device average · llama.cpp · MTP
unsloth/gemma-4-E2B-it-qat-GGUF gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
79.9Tater score
Limited — Not Ready Required accuracy 84.0% is below 85%. Routing 72.8% is below 80%.
82.3 raw formula84.0% required accuracy181.02 tok/s0.09s TTFT2.2 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 72.79%Spudex (optional) 73.41%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:52:46.830570+00:00 2 Limited — Not Ready 79.9 82.3 84.0% 181.02 0.09s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 75.6 accuracy + 4.8 speed + 1.5 TTFT + 0.4 memory + -2.4 readiness gate
#6 · Device average · llama.cpp · BASELINE
TaterTotterson/Qwen3.8-27B-GGUF-Tater-NoThink Qwen3.8-27B-Tater-NoThink-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
49.9Tater score
Not Fit Required accuracy 55.0% is below 70%. Routing 9.4% is below 70%.
50.2 raw formula55.0% required accuracy11.03 tok/s2.70s TTFT2.4 GiB peak RSS1 device1 unique run2 observations
Chat 84.29%Routing 9.38%Spudex (optional) 25.27%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:46:35.898683+00:00 2 Not Fit 49.9 50.2 55.0% 11.03 2.70s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 49.5 accuracy + 0.3 speed + 0.1 TTFT + 0.4 memory + -0.3 readiness gate
#7 · Device average · llama.cpp · BASELINE
unsloth/gemma-4-12B-it-qat-GGUF gemma-4-12B-it-qat-UD-Q4_K_XL.gguf
Vision Video Audio Declared model inputs · not scored
41.8Tater score
Not Fit Required accuracy 45.1% is below 70%. Routing 1.6% is below 70%.
41.8 raw formula45.1% required accuracy26.70 tok/s0.94s TTFT2.3 GiB peak RSS1 device1 unique run2 observations
Chat 60.71%Routing 1.62%Spudex (optional) 0.00%Synthesis 90.00%Tool Accuracy 78.00%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:52:23.303769+00:00 2 Not Fit 41.8 41.8 45.1% 26.70 0.94s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 40.6 accuracy + 0.7 speed + 0.1 TTFT + 0.4 memory + 0.0 readiness gate
#8 · Device average · llama.cpp · BASELINE
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
40.2Tater score
Not Fit Required accuracy 42.3% is below 70%. Routing 12.8% is below 70%.
40.2 raw formula42.3% required accuracy52.29 tok/s0.92s TTFT1.5 GiB peak RSS1 device1 unique run2 observations
Chat 84.29%Routing 12.75%Spudex (optional) 25.27%Synthesis 35.00%Tool Accuracy 78.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:41:02.780148+00:00 2 Not Fit 40.2 40.2 42.3% 52.29 0.92s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 38.1 accuracy + 1.4 speed + 0.1 TTFT + 0.6 memory + 0.0 readiness gate
#9 · Device average · llama.cpp · BASELINE
TaterTotterson/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF-Tater-NoThink NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Tater-NoThink-Q4_0.gguf
Vision Video Declared model inputs · not scored
36.2Tater score
Not Fit Required accuracy 37.4% is below 70%. Tool accuracy 58.3% is below 70%. Routing 0.0% is below 70%.
36.2 raw formula37.4% required accuracy60.41 tok/s0.97s TTFT1.1 GiB peak RSS1 device1 unique run2 observations
Chat 52.86%Routing 0.00%Spudex (optional) 22.53%Synthesis 90.00%Tool Accuracy 58.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:39:35.268844+00:00 2 Not Fit 36.2 36.2 37.4% 60.41 0.97s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 33.6 accuracy + 1.6 speed + 0.1 TTFT + 0.8 memory + 0.0 readiness gate
#10 · Device average · llama.cpp · BASELINE
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
35.7Tater score
Not Fit Required accuracy 37.7% is below 70%. Tool accuracy 47.3% is below 70%. Routing 7.9% is below 70%.
35.7 raw formula37.6% required accuracy44.12 tok/s0.65s TTFT2.1 GiB peak RSS1 device1 unique run2 observations
Chat 60.71%Routing 7.88%Spudex (optional) 0.00%Synthesis 90.00%Tool Accuracy 47.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:36:18.796223+00:00 2 Not Fit 35.7 35.7 37.6% 44.12 0.65s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 33.9 accuracy + 1.2 speed + 0.2 TTFT + 0.4 memory + 0.0 readiness gate
#11 · Device average · llama.cpp · BASELINE
allenai/OLMoE-1B-7B-0924-Instruct-GGUF olmoe-1b-7b-0924-instruct-q4_k_m.gguf
Vision Video Declared model inputs · not scored
29.3Tater score
Not Fit Required accuracy 25.0% is below 70%. Tool accuracy 16.7% is below 70%. Routing 0.0% is below 70%. Required accuracy 25.8% is below 70%. Tool accuracy 19.2% is below 70%.
29.3 raw formula25.4% required accuracy164.86 tok/s0.09s TTFT1.4 GiB peak RSS1 device2 unique runs2 observations
Chat 50.00%Routing 0.00%Spudex (optional) 0.00%Synthesis 100.00%Tool Accuracy 17.92%
See 2 unique runs Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:35:14.066851+00:00 1 Not Fit — Provisional 28.9 28.9 25.0% 164.91 0.09s AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:15:29.848645+00:00 1 Not Fit — Provisional 29.7 29.7 25.8% 164.81 0.09s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 22.9 accuracy + 4.4 speed + 1.4 TTFT + 0.7 memory + 0.0 readiness gate
#12 · Device average · llama.cpp · BASELINE
unsloth/Qwen3.5-0.8B-GGUF Qwen3.5-0.8B-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
28.6Tater score
Not Fit Required accuracy 24.3% is below 70%. Tool accuracy 32.0% is below 70%. Routing 0.0% is below 70%.
28.6 raw formula24.3% required accuracy174.76 tok/s0.12s TTFT917.4 MiB peak RSS1 device1 unique run2 observations
Chat 57.86%Routing 0.00%Spudex (optional) 0.00%Synthesis 62.50%Tool Accuracy 32.00%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:53:24.967361+00:00 2 Not Fit 28.6 28.6 24.3% 174.76 0.12s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 21.9 accuracy + 4.7 speed + 1.1 TTFT + 1.0 memory + 0.0 readiness gate
#13 · Device average · llama.cpp · BASELINE
unsloth/gemma-4-E4B-it-qat-GGUF gemma-4-E4B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
26.5Tater score
Not Fit Required accuracy 26.2% is below 70%. Tool accuracy 21.7% is below 70%. Routing 1.6% is below 70%. Required accuracy 26.3% is below 70%.
26.5 raw formula26.2% required accuracy58.06 tok/s0.12s TTFT2.5 GiB peak RSS1 device2 unique runs2 observations
Chat 59.28%Routing 1.62%Spudex (optional) 0.00%Synthesis 90.00%Tool Accuracy 21.67%
See 2 unique runs Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:53:12.950872+00:00 1 Not Fit — Provisional 26.4 26.4 26.2% 58.08 0.12s AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:33:33.305318+00:00 1 Not Fit — Provisional 26.6 26.6 26.3% 58.05 0.12s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 23.6 accuracy + 1.6 speed + 1.0 TTFT + 0.3 memory + 0.0 readiness gate
#14 · Device average · llama.cpp · BASELINE
unsloth/gemma-4-E2B-it-qat-GGUF gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
17.5Tater score
Not Fit Required accuracy 14.3% is below 70%. Tool accuracy 16.7% is below 70%. Routing 0.0% is below 70%.
17.5 raw formula14.3% required accuracy98.81 tok/s0.08s TTFT2.0 GiB peak RSS1 device1 unique run2 observations
Chat 52.86%Routing 0.00%Spudex (optional) 0.00%Synthesis 35.00%Tool Accuracy 16.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT AMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 64.0 GiB 2026-08-22T18:52:34.762990+00:00 2 Not Fit 17.5 17.5 14.3% 98.81 0.08s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 12.9 accuracy + 2.6 speed + 1.5 TTFT + 0.4 memory + 0.0 readiness gate
#1 · Device average · mlx · BASELINE
TaterTotterson/Gemma-4-26B-A4B-IT-UD-Q4_K_XL-mlx-Tater-NoThink 674e8a504329b470c761c2e1feeb31d93d6ceeb7
Text only Declared model inputs · not scored
88.6Tater score
Tater Ready All Tater readiness gates pass.
88.6 raw formula96.1% required accuracy55.09 tok/s0.23s TTFT18.5 GiB peak RSS1 device1 unique run2 observations
Chat 97.14%Routing 95.42%Spudex (optional) 97.58%Synthesis 90.00%Tool Accuracy 100.00%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:41:01.118063+00:00 2 Tater Ready 88.6 88.6 96.1% 55.09 0.23s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 86.5 accuracy + 1.5 speed + 0.6 TTFT + 0.1 memory + 0.0 readiness gate
#2 · Device average · llama.cpp · MTP
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
86.8Tater score
Tater Ready All Tater readiness gates pass.
86.8 raw formula93.0% required accuracy104.57 tok/s0.54s TTFT18.9 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:29:18.352445+00:00 2 Tater Ready 86.8 86.8 93.0% 104.57 0.54s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 2.8 speed + 0.2 TTFT + 0.1 memory + 0.0 readiness gate
#3 · Device average · llama.cpp · BASELINE
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
86.1Tater score
Tater Ready All Tater readiness gates pass.
86.1 raw formula93.0% required accuracy79.73 tok/s0.53s TTFT18.2 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:28:50.368818+00:00 2 Tater Ready 86.1 86.1 93.0% 79.73 0.53s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 2.1 speed + 0.2 TTFT + 0.1 memory + 0.0 readiness gate
#4 · Device average · llama.cpp · DFLASH
TaterTotterson/gemma-4-26B-A4B-it-GGUF-Tater-NoThink gemma-4-26B-A4B-it-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
85.0Tater score
Tater Ready All Tater readiness gates pass.
85.0 raw formula93.0% required accuracy39.13 tok/s0.54s TTFT21.3 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 93.00%Spudex (optional) 90.55%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:30:03.091430+00:00 2 Tater Ready 85.0 85.0 93.0% 39.13 0.54s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 83.7 accuracy + 1.0 speed + 0.2 TTFT + 0.0 memory + 0.0 readiness gate
#5 · Device average · llama.cpp · BASELINE
unsloth/gemma-4-12B-it-qat-GGUF gemma-4-12B-it-qat-UD-Q4_K_XL.gguf
Vision Video Audio Declared model inputs · not scored
84.4Tater score
Tater Ready All Tater readiness gates pass.
84.4 raw formula91.7% required accuracy62.96 tok/s1.22s TTFT8.5 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 90.08%Spudex (optional) 77.86%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:39:32.847663+00:00 2 Tater Ready 84.4 84.4 91.7% 62.96 1.22s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 82.5 accuracy + 1.7 speed + 0.1 TTFT + 0.1 memory + 0.0 readiness gate
#6 · Device average · llama.cpp · BASELINE
TaterTotterson/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF-Tater-NoThink NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Tater-NoThink-Q4_0.gguf
Vision Video Declared model inputs · not scored
84.4Tater score
Tater Ready All Tater readiness gates pass.
84.4 raw formula90.6% required accuracy97.59 tok/s0.66s TTFT16.5 GiB peak RSS1 device1 unique run2 observations
Chat 97.14%Routing 93.00%Spudex (optional) 82.09%Synthesis 90.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:30:52.427749+00:00 2 Tater Ready 84.4 84.4 90.6% 97.59 0.66s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 81.6 accuracy + 2.6 speed + 0.2 TTFT + 0.1 memory + 0.0 readiness gate
#7 · Device average · llama.cpp · MTP
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.4Tater score
Tater Ready All Tater readiness gates pass.
82.4 raw formula88.0% required accuracy111.59 tok/s0.61s TTFT23.1 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 83.00%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:32:27.413758+00:00 2 Tater Ready 82.4 82.4 88.0% 111.59 0.61s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 79.2 accuracy + 3.0 speed + 0.2 TTFT + 0.0 memory + 0.0 readiness gate
#8 · Device average · llama.cpp · BASELINE
TaterTotterson/Qwen3.8-27B-GGUF-Tater-NoThink Qwen3.8-27B-Tater-NoThink-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.2Tater score
Tater Ready All Tater readiness gates pass.
82.2 raw formula90.4% required accuracy27.18 tok/s3.30s TTFT17.5 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 88.42%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:35:34.657738+00:00 2 Tater Ready 82.2 82.2 90.4% 27.18 3.30s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 81.4 accuracy + 0.7 speed + 0.0 TTFT + 0.1 memory + 0.0 readiness gate
#9 · Device average · llama.cpp · MTP
TaterTotterson/Qwen3.8-27B-GGUF-Tater-NoThink Qwen3.8-27B-Tater-NoThink-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
82.1Tater score
Tater Ready All Tater readiness gates pass.
82.1 raw formula90.4% required accuracy25.24 tok/s3.34s TTFT23.8 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 88.42%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:38:40.509931+00:00 2 Tater Ready 82.1 82.1 90.4% 25.24 3.34s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 81.4 accuracy + 0.7 speed + 0.0 TTFT + 0.0 memory + 0.0 readiness gate
#10 · Device average · llama.cpp · BASELINE
TaterTotterson/Qwen3.6-35B-A3B-MTP-GGUF-Tater-NoThink Qwen3.6-35B-A3B-UD-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
81.5Tater score
Tater Ready All Tater readiness gates pass.
81.5 raw formula88.0% required accuracy75.91 tok/s0.61s TTFT20.4 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 83.00%Spudex (optional) 90.55%Synthesis 100.00%Tool Accuracy 86.67%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:31:46.399367+00:00 2 Tater Ready 81.5 81.5 88.0% 75.91 0.61s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 79.2 accuracy + 2.0 speed + 0.2 TTFT + 0.0 memory + 0.0 readiness gate
#11 · Device average · llama.cpp · BASELINE
unsloth/gemma-4-E4B-it-qat-GGUF gemma-4-E4B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
79.9Tater score
Limited — Not Ready Routing 75.4% is below 80%.
80.3 raw formula85.2% required accuracy99.13 tok/s0.16s TTFT5.4 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 75.42%Spudex (optional) 87.31%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:40:17.672114+00:00 2 Limited — Not Ready 79.9 80.3 85.2% 99.13 0.16s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 76.7 accuracy + 2.6 speed + 0.8 TTFT + 0.2 memory + -0.4 readiness gate
#12 · Device average · llama.cpp · BASELINE
unsloth/gemma-4-E2B-it-qat-GGUF gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
79.9Tater score
Limited — Not Ready Required accuracy 83.5% is below 85%. Routing 71.7% is below 80%.
80.6 raw formula83.5% required accuracy141.61 tok/s0.09s TTFT3.7 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 71.67%Spudex (optional) 73.41%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:39:45.924912+00:00 2 Limited — Not Ready 79.9 80.6 83.5% 141.61 0.09s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 75.2 accuracy + 3.8 speed + 1.4 TTFT + 0.2 memory + -0.7 readiness gate
#13 · Device average · llama.cpp · MTP
unsloth/gemma-4-E2B-it-qat-GGUF gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf
Vision Video Declared model inputs · not scored
79.9Tater score
Limited — Not Ready Required accuracy 83.5% is below 85%. Routing 71.7% is below 80%.
81.1 raw formula83.5% required accuracy160.75 tok/s0.09s TTFT3.9 GiB peak RSS1 device1 unique run2 observations
Chat 100.00%Routing 71.67%Spudex (optional) 73.41%Synthesis 90.00%Tool Accuracy 93.33%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:39:57.844280+00:00 2 Limited — Not Ready 79.9 81.1 83.5% 160.75 0.09s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 75.2 accuracy + 4.3 speed + 1.4 TTFT + 0.2 memory + -1.2 readiness gate
#14 · Device average · llama.cpp · BASELINE
unsloth/Qwen3.5-0.8B-GGUF Qwen3.5-0.8B-Q4_K_M.gguf
Vision Video Declared model inputs · not scored
49.9Tater score
Not Fit Required accuracy 67.3% is below 70%. Tool accuracy 62.2% is below 70%. Routing 59.3% is below 70%.
68.6 raw formula67.3% required accuracy229.34 tok/s0.12s TTFT1.2 GiB peak RSS1 device1 unique run2 observations
Chat 92.14%Routing 59.29%Spudex (optional) 39.51%Synthesis 90.83%Tool Accuracy 62.22%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:40:27.739073+00:00 2 Not Fit 49.9 68.6 67.3% 229.34 0.12s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 60.6 accuracy + 6.1 speed + 1.1 TTFT + 0.8 memory + -18.7 readiness gate
#15 · Device average · llama.cpp · BASELINE
allenai/OLMoE-1B-7B-0924-Instruct-GGUF olmoe-1b-7b-0924-instruct-q4_k_m.gguf
Vision Video Declared model inputs · not scored
46.3Tater score
Not Fit Required accuracy 41.3% is below 70%. Tool accuracy 44.8% is below 70%. Routing 11.6% is below 70%.
46.3 raw formula41.3% required accuracy262.69 tok/s0.06s TTFT5.8 GiB peak RSS1 device1 unique run2 observations
Chat 81.43%Routing 11.62%Spudex (optional) 25.38%Synthesis 100.00%Tool Accuracy 44.78%
See 1 unique run Device Finished Observations Fitness Final score Raw formula Required accuracy tok/s TTFT Apple M3 Ultra · 96.0 GiB 2026-08-22T18:28:14.739220+00:00 2 Not Fit 46.3 46.3 41.3% 262.69 0.06s
Suite tater-core-0.4 · Prompt tater-verba-nl-synthetic-2026-08-22 · Score mix: 37.1 accuracy + 7.0 speed + 2.0 TTFT + 0.1 memory + 0.0 readiness gate
