Machine
Apple M5 Max · Apple M5 Max #5e3bf11b
100 runs of 40 models since Sep 27, 2026. Last run Nov 20, 2026.
100
runs
40
models
88
configs
10
configs run 2+ times
Hardware
- GPU
- Apple M5 Max 64 GB
- CPU
- Apple M5 Max 18 cores
- RAM
- 64 GB
- OS
- macOS 27.2.1
- Tools
- LM Studio 47 · oMLX 20 · Unsloth Studio 14 · splash 6 · llama.cpp 6 · vLLM 4 · OpenAI-compatible 3
- 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, 40 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
unsloth/Qwen3.8-27B-GGUF
3 runs · 3 configs
qwen/qwen3.8-27b
1 run · 1 config
Muse-Glimmer-30B-4bit
5 runs · 5 configs
qwen3.8-27b-splash
10 runs · 7 configs
★ best4bitLM Studio256k ctxthinking on
×2 runs
85.384.6–86.0
steady
73.4 tok/s
71.5–75.3
mlx-community/Qwen3.8-27B-4bit
10 runs · 7 configs
★ bestsplash128k ctxthinking default
×3 runs
85.283.9–85.4
steady
89.3 tok/s
88.5–101
Swift-1.5-Qwen3.8-27b-oQ8e-bf16-mtp-MLX
1 run · 1 config
Tiel-Coder-35B-A3B-MLX-oQ4e
2 runs · 2 configs
qwen3.8-27b-gsq-rco
1 run · 1 config
muse-glimmer-30b
3 runs · 2 configs
★ best4bitLM Studio128k ctxthinking default
×2 runs
84.082.4–85.6
steady
30.8 tok/s
30.6–31.1
Qwen3.8-27B-oQ4e-mtp
8 runs · 7 configs
★ bestoQ4eoMLX256k ctxthinking oneffort mediumKV none
×2 runs
83.681.7–85.5
steady
29.7 tok/s
29.5–29.9
Compare with another machine
Run history
Nov 19, 2026
23:29Qwen3.8-27B-4bitsplash101 tok/s85.2Nov 14, 2026
16:19Qwen3.8-27B-4bitsplash88.5 tok/s85.415:57flint-GGUF:UD-Q4_K_XLsplash201 tok/s71.415:40Qwen3.8-27B-4bitsplash89.3 tok/s83.915:18gemma-4-31B-it-qat-GGUFUnsloth Studio50.8 tok/s76.215:14gemma-4-31B-it-qat-GGUFUnsloth Studio50.8 tok/s72.614:51Qwen3.8-27B-GGUFUnsloth Studio35.2 tok/s86.814:24gemma-4-31B-it-qat-GGUFUnsloth Studio50.1 tok/s73.914:19gemma-4-31B-it-qat-GGUFUnsloth Studio50.5 tok/s77.014:14Muse-Glimmer-30B-4bitUnsloth Studio30.9 tok/s81.214:05Muse-Glimmer-30B-4bitUnsloth Studio30.9 tok/s83.013:44gemma-4-31b-it-UD-MLX-4bitUnsloth Studio21.4 tok/s76.413:40gemma-4-31b-it-UD-MLX-4bitUnsloth Studio21.7 tok/s73.4Nov 6, 2026
21:10gemma-4-e4bLM Studio96.6 tok/s59.621:01Swift-1.5-Qwen3.8-27b-oQ8e-bf16-mtp-MLXoMLX47.0 tok/s85.004:06Qwen3.8-27B-oQ4e-mtpoMLX28.4 tok/s87.003:20Qwen3.8-27B-4bitOpenAI-compatible84.4 tok/s86.902:22nanbeige4.3-33bLM Studio19.5 tok/s43.1Nov 5, 2026
03:50twil-lm3-proLM Studio62.2 tok/s60.0Show all 100 runs ▾Show fewer ▴
Nov 4, 2026
23:27LFM3-1.7B-MLX-bf16oMLX126 tok/s64.2Nov 1, 2026
17:41Muse-Glimmer-30B-4bitoMLX30.9 tok/s86.317:30unsloth--gemma-4-31b-it-UD-MLX-4bitoMLX16.2 tok/s73.4Oct 31, 2026
17:32muse-glimmer-30bLM Studio31.1 tok/s82.417:22qwen3-coder-nextLM Studio76.9 tok/s78.5Oct 30, 2026
23:18gemma-4-31bLM Studio15.2 tok/s78.123:04gemma-4-31bLM Studio15.1 tok/s75.422:15gemma-4-26b-a4bLM Studio113 tok/s67.322:13gemma-4-26b-a4bLM Studio113 tok/s68.122:05gemma-4-26b-a4bLM Studio113 tok/s73.822:00flintLM Studio109 tok/s68.021:55flintLM Studio109 tok/s73.811:00flint×2LM Studio106 tok/s77.702:59bonsai-27bLM Studio52.3 tok/s69.6Oct 29, 2026
22:36falcon3-10b-instructLM Studio46.9 tok/s49.522:22gemma-4-26b-a4bLM Studio118 tok/s73.4Oct 21, 2026
23:28Qwen3.8-27B-4bitvLLM74.9 tok/s83.0Oct 17, 2026
14:05Qwen3.8-27B-4bitvLLM90.5 tok/s85.213:32Qwen3.8-27B-4bitvLLM84.0 tok/s87.712:48muse-glimmer-30bLM Studio30.6 tok/s85.6Oct 16, 2026
20:43Qwen3.8-27B-NVFP4OpenAI-compatible75.4 tok/s76.720:38Qwen3.8-27B-NVFP4OpenAI-compatible75.6 tok/s73.6Oct 15, 2026
23:42Tiel-Coder-35B-A3B-MLX-oQ4eoMLX88.0 tok/s82.123:30Tiel-Coder-35B-A3B-MLX-oQ4eoMLX54.5 tok/s84.723:04Qwen3.8-27B-GGUFllama.cpp29.4 tok/s87.422:45Qwen3.8-27B-GGUFllama.cpp29.5 tok/s86.1Oct 14, 2026
23:17qwen3.8-27b-splashLM Studio102 tok/s80.823:00qwen3.8-27b-splashLM Studio99.5 tok/s80.621:57Muse-Glimmer-30B-4bitoMLX31.3 tok/s78.701:13Muse-Glimmer-30B-GGUFllama.cpp17.3 tok/s82.500:35Muse-Glimmer-30B-GGUF:UD-Q4_K_XLllama.cpp30.8 tok/s82.000:10Muse-Glimmer-30B-4bitoMLX31.1 tok/s81.6Oct 13, 2026
23:50muse-glimmer-30bLM Studio30.9 tok/s76.9Oct 11, 2026
22:18nvidia-nemotron-3.5-lightning-30b-a3bLM Studio109 tok/s76.422:07nvidia-nemotron-3.5-lightning-30b-a3bLM Studio112 tok/s74.521:55nvidia-nemotron-3.5-lightning-30b-a3bLM Studio114 tok/s73.315:15nvidia-nemotron-3.5-lightning-30b-a3bLM Studio112 tok/s78.5Oct 10, 2026
17:44qwen3.8-27b-splashLM Studio76.3 tok/s84.717:23qwen3.8-27b-splashLM Studio74.4 tok/s84.314:44Qwen3.8-27B-4bitvLLM82.2 tok/s82.614:14qwen3.8-27b-splashLM Studio71.2 tok/s83.413:56Tiel-Coder-35B-A3B-GGUF-MTPUnsloth Studio101 tok/s81.013:29gemma-4-31b-it-UD-MLX-4bitUnsloth Studio20.9 tok/s75.813:17unsloth--gemma-4-31b-it-UD-MLX-4bitoMLX21.0 tok/s75.713:06gpt-oss-20b-MLX-8bitoMLX95.7 tok/s74.2Oct 9, 2026
20:42mellum2.1-12b-a2.5b-thinkingLM Studio96.2 tok/s68.820:27mellum2.1-12b-a2.5b-thinkingLM Studio98.3 tok/s64.120:20mellum2.1-12b-a2.5b-thinkingLM Studio96.8 tok/s70.902:39moonshotai_kimi-linear-48b-a3b-instructLM Studio107 tok/s60.502:38moonshotai_kimi-linear-48b-a3b-instructLM Studio111 tok/s62.401:54qwen3.8-27b-gsq-rcoLM Studio43.8 tok/s84.6Oct 3, 2026
19:49Qwen3.8-27B-oQ4e-mtpoMLX29.1 tok/s85.217:50Qwen3.8-27B-oQ4e-mtpoMLX29.1 tok/s84.517:11Qwen3.8-27B-oQ4e-mtpoMLX29.6 tok/s80.516:39qwen3.8-27b-splashLM Studio71.5 tok/s86.016:31qwen3.8-27b-splashLM Studio73.1 tok/s73.816:16gemma-4-31B-it-qat-GGUFllama.cpp28.8 tok/s74.616:07gemma-4-31B-it-GGUFllama.cpp33.2 tok/s70.8Oct 2, 2026
16:25qwen3.8-27b-splashLM Studio75.3 tok/s84.615:30Qwen3.8-35B-A3B-Distill-MLX-APEX-MiniPlus-AbliteratedoMLX49.1 tok/s75.515:05ThinkingCap-Qwen3.8-27B-MLX-4bit-DWQoMLX24.8 tok/s82.414:05Qwen3.8-27B-oQ4e-mtp×2oMLX29.7 tok/s83.602:47Qwen3.8-27B-oQ4e-mtpoMLX29.8 tok/s69.400:53Qwen3.8-27B-oQ4e-mtpoMLX28.9 tok/s84.5Oct 1, 2026
23:53GLM-4.7-Flash-GGUFUnsloth Studio73.9 tok/s70.023:32gemma-4-31b-it-UD-MLX-4bitUnsloth Studio21.6 tok/s78.822:59gemma-4-31B-it-qat-GGUFUnsloth Studio42.1 tok/s72.301:11qwen3.8-27bLM Studio17.8 tok/s86.4Sep 30, 2026
23:34gemma-4-31bLM Studio14.9 tok/s75.1