Machine
Apple M5 Max · Apple M5 Max #32527cfb
304 runs of 78 models since May 12, 2026. Last run Nov 21, 2026.
304
runs
78
models
126
configs
66
configs run 2+ times
Hardware
- GPU
- Apple M5 Max 64 GB
- CPU
- Apple M5 Max 18 cores
- RAM
- 64 GB
- OS
- macOS 26.6.2
- Tools
- oMLX 283 · Ollama 16 · llama.cpp 4 · LM Studio 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, 78 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-MTPLX-Optimized-Quality
2 runs · 1 config
★ bestoMLX256k ctxthinking oneffort xhighKV none
×2 runs
89.188.9–89.3
steady
34.3 tok/s
32.8–35.8
DaisyRose-27B-oQ8e-mtp
6 runs · 1 config
★ bestoQ8eoMLX256k ctxthinking oneffort xhighKV none
×6 runs
88.884.7–89.1
steady
50.0 tok/s
49.1–52.2
Qwen3.8-27B-oQ4e-mtp
30 runs · 12 configs
★ bestoMLX256k ctxthinking oneffort xhighKV none
×2 runs
87.386.7–88.0
steady
44.2 tok/s
41.2–47.1
Inkling-2-Mini-oQ4e-mtp
6 runs · 1 config
★ bestoQ4eoMLX256k ctxthinking onKV none
×6 runs
87.261.4–88.8
2 outliers
57.6 tok/s
54.1–59.6
Qwen3.8-27B-oQ8e-fp16-mtp
4 runs · 2 configs
★ bestFP16oMLX64k ctxthinking default
×2 runs
87.285.6–88.8
steady
35.4 tok/s
35.4–35.4
Qwen3.8-27B-oQ8e-mtp
62 runs · 16 configs
★ bestoQ8eoMLX170k ctxthinking oneffort xhighKV none
×7 runs
86.984.5–89.9
steady
51.7 tok/s
46.6–52.3
Qwen3.8-27B-Q8_0
2 runs · 1 config
★ bestQ8_0oMLX256k ctxthinking oneffort xhighKV none
×2 runs
86.585.6–87.3
steady
53.9 tok/s
53.3–54.6
Tiel-Coder-35B-A3B-MLX-oQ6e-MTP
1 run · 1 config
Swift-Qwen3.8-27b-oQ4e-fp16-mtp
2 runs · 1 config
★ bestFP16oMLX256k ctxthinking oneffort xhighKV none
×2 runs
86.185.5–86.7
steady
46.7 tok/s
44.8–48.6
Hemmingway-1-oQ8e-mtp
4 runs · 1 config
★ bestoQ8eoMLX256k ctxthinking oneffort xhighKV none
×4 runs
86.082.1–87.6
steady1 slow run
31.7 tok/s
17.5–32.0
Compare with another machine
Run history
Nov 21, 2026
18:32Inkling-2-Mini-oQ4e-mtp×6oMLX57.6 tok/s87.2Nov 20, 2026
19:20DaisyRose-27B-oQ8e-mtp×6oMLX50.0 tok/s88.8Oct 27, 2026
14:18Qwen3.8-27B-oQ8e-mtpoMLX37.6 tok/s89.013:41Qwen3.8-27B-oQ8e-mtpoMLX48.2 tok/s85.613:15Qwen3.8-27B-oQ8e-mtpoMLX37.6 tok/s90.112:42Qwen3.8-27B-oQ8e-mtpoMLX50.3 tok/s87.911:54Qwen3.8-27B-oQ8e-mtpoMLX39.8 tok/s87.411:47Qwen3.8-27B-oQ8e-mtpoMLX52.4 tok/s84.011:42Qwen3.8-27B-oQ8e-mtpoMLX40.5 tok/s82.711:34Qwen3.8-27B-oQ8e-mtpoMLX56.3 tok/s84.508:56Qwen3.8-27B-oQ8e-mtpoMLX44.9 tok/s74.508:49Qwen3.8-27B-oQ8e-mtpoMLX43.4 tok/s69.4Oct 26, 2026
21:11Qwen3.8-27B-oQ8e-mtpoMLX39.4 tok/s73.921:05Qwen3.8-27B-oQ8e-mtpoMLX51.6 tok/s79.421:01Qwen3.8-27B-oQ8e-mtpoMLX38.2 tok/s64.220:55Qwen3.8-27B-oQ8e-mtpoMLX55.0 tok/s74.320:44Qwen3.8-27B-oQ8e-mtpoMLX38.1 tok/s85.520:16Qwen3.8-27B-oQ8e-mtpoMLX49.0 tok/s86.919:50Qwen3.8-27B-oQ8e-mtpoMLX38.5 tok/s89.619:15Qwen3.8-27B-oQ8e-mtpoMLX50.8 tok/s83.018:30Qwen3.8-27B-oQ8e-mtpoMLX45.5 tok/s83.818:23Qwen3.8-27B-oQ8e-mtpoMLX61.4 tok/s70.7Show all 304 runs ▾Show fewer ▴
Oct 22, 2026
18:41Qwen3.8-27B-oQ8e-mtpoMLX51.8 tok/s88.8Oct 15, 2026
18:48LFM2.5-24B-A2B-MLX-8bit×5oMLX126 tok/s67.2Oct 6, 2026
20:49Qwen3.8-Flash-Next-oQ4e-mtpoMLX27.9 tok/s81.120:16Qwen3.8-27B-oQ8e-mtpoMLX47.7 tok/s69.9Oct 2, 2026
17:37Qwen3.8-27B-oQ8e-mtpoMLX39.7 tok/s82.617:11Tiel-Coder-35B-A3B-MLX-oQ6e-MTPoMLX167 tok/s86.3Sep 29, 2026
18:14Qwen3.8-27B-oQ8e-mtp×2oMLX34.7 tok/s86.416:15Hemmingway-1-oQ8e-mtp×4oMLX31.7 tok/s86.011:00Swift-1.5-Qwen3.8-27b-oQ8e-mtp×4oMLX34.7 tok/s85.808:50Qwen3.8-Flash-Next-oQ4e-mtpoMLX23.7 tok/s84.8Sep 28, 2026
21:09Swift-1.5-Qwen3.8-27b-oQ8e-mtpoMLX18.0 tok/s83.020:29Qwen3.8-Flash-Next-oQ4e-mtpoMLX23.6 tok/s77.0Sep 26, 2026
21:36Qwen3.8-27B-oQ8e-mtpoMLX32.9 tok/s85.7Sep 24, 2026
11:15Swift-Qwen3.8-27b-oQ4e-fp16-mtp×2oMLX46.7 tok/s86.109:07granite-4.2-30b-MLX-8bit×2oMLX16.0 tok/s72.1Sep 21, 2026
19:47Qwen3.8-27B-oQ4e-mtpoMLX37.9 tok/s83.919:14Qwen3.8-27B-oQ4e-mtpoMLX37.3 tok/s85.018:38Qwen3.8-27B-oQ4e-mtpoMLX39.5 tok/s85.817:55Qwen3.8-27B-oQ4e-mtpoMLX44.1 tok/s85.6Sep 13, 2026
18:38Agnes-3.0-Flash-MLX-4bitoMLX23.7 tok/s82.3Sep 12, 2026
10:27Nex-N2.5-mini-MLX-4bit×15oMLX137 tok/s81.306:33Nex-N2.5-mini-MLX-4bit×5oMLX138 tok/s80.7Sep 11, 2026
22:32Qwen3.8-27B-MTPLX-Optimized-QualityoMLX32.8 tok/s88.920:15Ornith-1.5-35B-A3B-oQ4e-mtpoMLX112 tok/s86.720:03Qwen3.8-27B-MTPLX-Optimized-QualityoMLX35.8 tok/s89.318:30K2-Horizon-MoVA-36B-A4B-MLX-4bitoMLX47.8 tok/s82.3Sep 10, 2026
05:36Qwen3.8-27B-oQ8e-mtpoMLX39.0 tok/s81.2Sep 9, 2026
20:44Qwen3.8-27B-oQ8e-mtpoMLX39.4 tok/s78.718:52Qwen3.8-27B-oQ4e-mtpoMLX47.1 tok/s88.018:24Qwen3.8-27B-oQ4e-mtp×3oMLX53.2 tok/s83.518:05Qwen3.8-27B-oQ4e-mtp×4oMLX53.9 tok/s83.606:02Qwen3.8-27B-oQ8e-mtpoMLX32.1 tok/s81.8Sep 8, 2026
21:08Qwen3.8-27B-oQ4e-mtpoMLX41.2 tok/s86.7Sep 7, 2026
20:49Qwen3.8-27B-oQ4e-mtp×4oMLX40.6 tok/s84.014:01Ornith-1.5-35B-A3B-oQ8e-mtpoMLX91.8 tok/s85.012:57Tiel-Coder-35B-A3B-MLX-oQ4e-MTPoMLX124 tok/s83.212:48Qwen3.6-35B-A3B-oQ8e-mtpoMLX98.6 tok/s80.2Sep 3, 2026
23:15Qwen3.6-27B-oQ8e-mtpoMLX35.0 tok/s81.123:05Qwen3.8-27B-oQ8e-mtpoMLX29.3 tok/s87.722:19Qwen3.8-27B-oQ8e-mtpoMLX27.5 tok/s86.321:40Qwen3.8-27B-oQ8e-mtpoMLX32.1 tok/s83.7Aug 30, 2026
18:58Qwen3.8-27B-oQ3e-mtpoMLX41.6 tok/s83.017:20Qwen3.8-27B-oQ2e-mtpoMLX46.3 tok/s10.516:33Qwen3.8-27B-oQ2e-mtpoMLX45.2 tok/s16.914:45Qwen3.8-27B-oQ8e-mtpoMLX27.3 tok/s87.013:45Qwen3.8-27B-oQ6e-mtpoMLX29.2 tok/s85.713:06Qwen3.8-27B-oQ4e-mtpoMLX36.7 tok/s86.812:31Qwen3.8-27B-oQ3e-mtpoMLX38.9 tok/s85.211:56Qwen3.8-27B-oQ2e-mtpoMLX40.6 tok/s3.611:11Qwen3.8-27B-oQ8e-mtpoMLX35.4 tok/s69.2Aug 29, 2026
22:25Qwen3.8-27B-oQ8e-mtpoMLX34.4 tok/s85.818:30Qwen3.8-27B-oQ8e-mtp×4oMLX38.5 tok/s73.816:56Qwen3.8-27B-oQ8e-mtpoMLX33.5 tok/s85.915:28Ornith-1.5-9B-oQ4e-mtpoMLX84.8 tok/s76.715:09Ornith-1.5-9B-MLX-8bitoMLX54.4 tok/s65.014:21Muse-Glimmer-30B-oQ8eoMLX17.2 tok/s84.313:49Ornith-1.5-35B-A3B-oQ8e-mtpoMLX98.3 tok/s84.410:42Qwen3.8-27B-oQ8e-mtp×4oMLX33.5 tok/s84.8Aug 28, 2026
23:47Qwen3.8-27B-oQ4-mtpoMLX46.1 tok/s87.613:06Qwen3.8-27B-oQ4e-mtpoMLX33.6 tok/s84.012:33Qwen3.8-27B-oQ8e-mtpoMLX30.0 tok/s83.911:38Qwen3.8-27B-oQ8e-fp16-mtpoMLX29.2 tok/s76.8Aug 27, 2026
19:15Qwen3.8-27B-oQ8e-mtpoMLX27.1 tok/s68.718:32Qwen3.8-27B-oQ4e-mtpoMLX44.9 tok/s87.316:57Qwen3.8-27B-oQ8e-mtpoMLX30.8 tok/s84.7Aug 26, 2026
20:29Qwen3.8-27B-oQ8e-mtp×2oMLX32.5 tok/s85.806:26Qwen3.8-27B-oQ4e-mtpoMLX47.0 tok/s69.4Aug 25, 2026
21:13Ornith-1.5-35B-A3B-oQ4e-mtpoMLX110 tok/s85.720:59Qwen3.8-27B-oQ4e-mtpoMLX39.9 tok/s83.420:14Tiel-Coder-35B-A3B-MLX-oQ4e-MTP×4oMLX124 tok/s82.117:23Tiel-Coder-35B-A3B-MLX-oQ4eoMLX119 tok/s84.316:45Qwen3.8-27B-oQ4e-mtpoMLX46.3 tok/s85.015:08Tiel-Coder-35B-A3B-MLX-oQ4e×3oMLX120 tok/s84.7Aug 24, 2026
18:23Ornith-1.5-9B-MLX-8bitoMLX51.7 tok/s62.116:23Ornith-1.5-9B-oQ8e×2oMLX53.0 tok/s65.609:51Laguna-XS-2.1-oQ8eoMLX100 tok/s62.509:23Qwen3.6-35B-A3B-oQ8e-mtpoMLX93.1 tok/s81.4Aug 23, 2026
23:25Muse-Glimmer-30B-oQ8eoMLX16.8 tok/s83.223:06Ornith-1.5-35B-A3B-oQ8e-mtpoMLX94.7 tok/s85.122:58NVIDIA-Nemotron-3.5-Lightning-30B-A3B-mlx-oQ8eoMLX107 tok/s77.322:46Qwen3.8-27B-oQ8e-mtpoMLX32.4 tok/s89.011:41Qwen3.8-27B-oQ4e-mtpoMLX43.0 tok/s85.909:59Muse-Glimmer-30B-oQ8eoMLX16.9 tok/s84.309:42Ornith-1.5-35B-A3B-oQ8e-mtpoMLX95.6 tok/s78.909:18NVIDIA-Nemotron-3.5-Lightning-30B-A3B-mlx-oQ8eoMLX106 tok/s78.4Aug 22, 2026
23:23Qwen3.8-27B-oQ4e-mtpoMLX42.2 tok/s87.222:53NVIDIA-Nemotron-3.5-Lightning-30B-A3B-oQ4e-mtpoMLX110 tok/s76.122:40Ornith-1.5-35B-A3B-oQ4e-mtpoMLX113 tok/s85.021:49Muse-Glimmer-30B-oQ4eoMLX28.7 tok/s82.319:15Qwen3.8-27B-oQ4e-mtp×2oMLX37.2 tok/s84.617:56Qwen3.6-27B-MLX-8bitoMLX15.3 tok/s80.215:56Qwen3.8-27B-4bitoMLX30.5 tok/s87.115:00Qwen3.8-27B-oQ8e-mtpoMLX33.9 tok/s82.514:36Qwen3.8-27B-oQ8e-fp16-mtpoMLX35.0 tok/s87.113:12Ornith-1.5-35B-A3B-MLX-8bitoMLX101 tok/s81.4Aug 21, 2026
19:30Qwen3.8-27B-Uncensored-MLXoMLX29.6 tok/s85.813:58Qwen3.8-27B-GGUF:UD-Q8_K_XLllama.cpp23.2 tok/s86.113:32Qwen3.8-27B-GGUF:UD-Q6_K_XLllama.cpp25.2 tok/s87.010:42Qwen3.8-27B-GGUF:UD-Q8_K_XLllama.cpp26.2 tok/s79.710:24Qwen3.8-27B-GGUF:UD-Q8_K_XLllama.cpp23.8 tok/s76.9Aug 20, 2026
12:41Qwen3.8-27B-oQ4e-mtpoMLX54.5 tok/s71.4Aug 19, 2026
21:56Ornith-1.5-35B-A3B-oQ8e-mtpoMLX92.7 tok/s85.221:38Ornith-1.5-35B-A3B-MLX-8bitoMLX92.6 tok/s82.520:15Muse-Glimmer-30B-oQ8e-fp16oMLX17.4 tok/s77.309:25Qwen3.8-27B-oQ8e-fp16-mtp×2oMLX35.4 tok/s87.206:35Qwen3.8-27B-oQ4e-mtpoMLX49.3 tok/s82.2Aug 18, 2026
21:09Qwen3.8-27B-oQ4e-mtpoMLX41.2 tok/s88.220:21Qwen3.8-27B-oQ8e-mtpoMLX32.3 tok/s84.819:56Qwen3.8-27B-4bitoMLX30.6 tok/s84.6Aug 17, 2026
21:03Qwen3.8-27B-4bitoMLX29.2 tok/s85.719:50Qwen3.8-27B-oQ4-mtp×2oMLX38.6 tok/s83.619:10Qwen3.8-27B-oQ8-mtp×2oMLX33.6 tok/s77.4Aug 14, 2026
23:04Qwen3.8-27B-4bit×4oMLX30.5 tok/s84.019:48Qwen3.8-27B-4bitoMLX30.8 tok/s78.219:36Qwen3.8-27B-8bitoMLX17.8 tok/s68.119:19Qwen3.8-27B-4bitoMLX31.9 tok/s76.011:45gemma-4-31B-it-MLX-8bitoMLX14.9 tok/s70.907:22Hermes-4-70B-MLX-4bitoMLX12.9 tok/s64.707:05granite-4.1-30b-mxfp8oMLX17.5 tok/s63.106:56Laguna-S-2.1-oQ3eoMLX71.6 tok/s68.0Aug 13, 2026
20:46Laguna-XS-2.1-NVFP4-mlxoMLX85.4 tok/s65.218:19Muse-Glimmer-30B-8bitoMLX17.2 tok/s81.817:42Muse-Glimmer-30B-4bit×3oMLX29.9 tok/s80.5Aug 12, 2026
21:10DeepSeek-V4-Pro-Qwen3.5-9BoMLX30.6 tok/s67.720:44Ornith-1.0-35B-MTPLXoMLX124 tok/s81.420:41LFM2.5-8B-A1B-MLX-8bit×2oMLX193 tok/s48.020:28KAT-Coder-V2.5-Dev-oQ8eoMLX99.9 tok/s82.820:05NVIDIA-Nemotron-3.5-Lightning-30B-A3B-8bitoMLX96.7 tok/s81.119:58Qwen3-Coder-30B-A3B-Instruct-MLX-8bit×2oMLX89.8 tok/s48.719:48Qwen3.5-27B-Claude-4.6-Opus-Distilled-MLX-4bitoMLX30.1 tok/s82.819:42Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking-8bitoMLX12.0 tok/s76.507:41NVIDIA-Nemotron-3.5-Lightning-30B-A3B-8bitoMLX110 tok/s75.5Aug 10, 2026
19:55gemma-4-31B-it-MLX-8bitoMLX14.9 tok/s75.019:19Qwen3.6-35B-A3B-MLX-8bitoMLX99.4 tok/s86.018:22Qwen3.6-27B-MLX-4bitoMLX32.1 tok/s83.312:30Qwen3.6-27B-MLX-8bit×2oMLX15.2 tok/s85.2Jul 17, 2026
20:10Qwen3.6:35B-4bitoMLX103 tok/s79.520:05Qwen3.6-35B-A3B-MLX-8bitoMLX90.1 tok/s83.119:59Qwen3.6-27B-MLX-4bitoMLX30.8 tok/s85.719:48Qwen3.6-27B-MLX-8bitoMLX15.3 tok/s81.319:27Qwen3.6-27B-MTPLX-Optimized-SpeedoMLX31.2 tok/s85.619:16Qwen3.6-27B-UD-MLX-4bit×2oMLX13.0 tok/s74.614:05Qwen3.6-27B-MTPLX-Optimized-SpeedoMLX7.4 tok/s77.113:40Qwen3.6-27B-MLX-4bitoMLX7.5 tok/s83.613:30Qwen3.6-27B-MLX-8bitoMLX4.2 tok/s77.7Jul 15, 2026
16:20Qwen3.6-27B-MLX-8bitoMLX3.9 tok/s81.908:32Qwen3.6:35B-4bitoMLX25.7 tok/s82.208:26Qwen3.6-27B-UD-MLX-4bitoMLX4.7 tok/s80.408:06Qwen3.6-35B-A3B-MLX-8bitoMLX24.7 tok/s80.207:40Qwen3.6-27B-MLX-8bitoMLX4.1 tok/s78.907:21Qwen3.6-27B-MLX-4bitoMLX8.4 tok/s81.305:45Qwen3.6-27B-MTPLX-Optimized-SpeedoMLX19.5 tok/s58.3Jul 14, 2026
22:52Qwen3.6-27B-MTPLX-Optimized-SpeedoMLX17.4 tok/s57.221:06Qwen3.6-27B-MLX-8bitoMLX11.3 tok/s78.719:53Qwen3.6-35B-A3B-MLX-8bitoMLX62.6 tok/s74.619:47Qwen3.6-27B-UD-MLX-4bitoMLX10.2 tok/s59.219:23Qwen3.6-27B-MLX-8bitoMLX10.7 tok/s81.219:01Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-Thinking-8bitoMLX8.3 tok/s73.8Jul 13, 2026
21:20Qwen3.6-27B-MTPLX-Optimized-SpeedoMLX21.9 tok/s84.320:07Qwen3.6-27B-MLX-6bitoMLX15.4 tok/s81.819:53Qwen3.6-27B-MLX-4bitoMLX22.3 tok/s80.718:16NVIDIA-Nemotron-3-Super-120B-A12B-OptiQ-2bitoMLX32.5 tok/s47.918:01NVIDIA-Nemotron-3-Super-120B-A12B-OptiQ-2bitoMLX41.6 tok/s73.117:55NVIDIA-Nemotron-3-Super-120B-A12B-OptiQ-2bitoMLX37.4 tok/s72.114:39Ornith-1.0-35B-MTPLXoMLX85.0 tok/s79.2Jul 12, 2026
17:59Qwen3.6-27B-UD-MLX-4bitoMLX14.6 tok/s78.6Jun 29, 2026
16:04gemma-4-12b-coder-fable5-composer2.5-8bitoMLX23.0 tok/s2.514:15Qwen3.6-27B-UD-MLX-4bitoMLX14.3 tok/s82.611:40Qwen3.6-35B-A3B-MLX-8bitoMLX69.2 tok/s77.710:56Ornith-1.0-35B-MLX-oQ8oMLX67.2 tok/s82.8Jun 2, 2026
21:31Qwen3.6-27B-UD-MLX-4bitoMLX15.0 tok/s76.819:42Qwen3.6-35B-A3B-MLX-8bitoMLX68.4 tok/s78.2May 24, 2026
18:18qwen3.6:27bOllama4.8 tok/s78.4May 23, 2026
21:10qwen3.6:27b-mlxOllama7.6 tok/s84.614:44gemma4:31b-mlxOllama7.7 tok/s72.613:25gemma4:31b-it-q8_0Ollama5.3 tok/s76.812:13gemma4:31b-mxfp8Ollama4.7 tok/s69.2May 21, 2026
06:55qwen3-coder-next:latestOllama28.9 tok/s78.206:40qwen3-coder-next:latestOllama33.4 tok/s75.106:27qwen3-coder-next:latestOllama34.1 tok/s74.2May 16, 2026
13:35qwen3.6:35bOllama19.6 tok/s77.012:53gemma4:31bOllama7.5 tok/s79.908:12qwen3.6:35bOllama15.0 tok/s77.8