Mode    Sampled         Qwen3.8-27B-Brainwaves-oQ4e-mtp  Qwen3.8-27B-Brainwaves-oQ6e-mtp  Qwen3.8-27B-Brainwaves-oQ8e-mtp           Qwen3.8-27B-Brainwaves         Qwen3.8-27B-MLX-oQ8e-mtp       Qwen3.6-27B-Seven-oQ8e-mtp
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
MMLU          Sample  300/14042                                 85.3%                            85.0%                            84.7%                            85.3%                            78.7%                            88.0%
MMLU_PRO      Sample  300/12032                                 69.3%                            69.3%                            69.7%                            69.0%                            65.7%                            70.7%
HUMANEVAL     Full    164                                       87.8%                            88.4%                            87.8%                            87.2%                            92.7%                            90.2%

--- Detail ---

Model: Qwen3.8-27B-Brainwaves-oQ4e-mtp
Benchmark         Accuracy   Correct   Total   Time(s)   Think
--------------------------------------------------------------
MMLU                 85.3%       256     300     328.4      No
MMLU_PRO             69.3%       208     300     151.2      No
HUMANEVAL            87.8%       144     164       241      No

Model: Qwen3.8-27B-Brainwaves-oQ6e-mtp
Benchmark         Accuracy   Correct   Total   Time(s)   Think
--------------------------------------------------------------
MMLU                 85.0%       255     300     363.7      No
MMLU_PRO             69.3%       208     300     172.5      No
HUMANEVAL            88.4%       145     164     296.5      No

Model: Qwen3.8-27B-Brainwaves-oQ8e-mtp
Benchmark         Accuracy   Correct   Total   Time(s)   Think
--------------------------------------------------------------
MMLU                 84.7%       254     300     374.7      No
MMLU_PRO             69.7%       209     300     176.1      No
HUMANEVAL            87.8%       144     164     316.3      No

Model: Qwen3.8-27B-Brainwaves
Benchmark         Accuracy   Correct   Total   Time(s)   Think
--------------------------------------------------------------
MMLU                 85.3%       256     300     402.4      No
MMLU_PRO             69.0%       207     300     232.3      No
HUMANEVAL            87.2%       143     164     825.2      No

Model: Qwen3.8-27B-MLX-oQ8e-mtp
Benchmark         Accuracy   Correct   Total   Time(s)   Think
--------------------------------------------------------------
MMLU                 78.7%       236     300     348.8      No
MMLU_PRO             65.7%       197     300     173.2      No
HUMANEVAL            92.7%       152     164       470      No

Model: Qwen3.6-27B-Seven-oQ8e-mtp
Benchmark         Accuracy   Correct   Total   Time(s)   Think
--------------------------------------------------------------
MMLU                 88.0%       264     300     370.3      No
MMLU_PRO             70.7%       212     300     178.4      No
HUMANEVAL            90.2%       148     164     334.4      No