Silicon Score Apple Silicon local LLM rankings
← Canonical rankings

Silicon Score › Chips › M5 Max (128 GB)

M5 Max (128 GB) — benchmark record

M5 Max (128 GB) local LLM benchmarks on Apple Silicon Mac with 33 published rows across 23 models. Peak published speed is 158.0 tok/s on Gemma 4 E2B. Largest published fit uses 92.6 GB. Published runtimes include flash-moe, llama.cpp, MLX, Ollama.

33Benchmark rows
23Models tested
158.0Fastest avg tok/s (Gemma 4 E2B)
0Lab benchmarks

Quick take

Best published speed here is 158.0 tok/s on Gemma 4 E2B at Q4_K - Medium. Largest published memory footprint is Qwen3-Coder-Next at 8bit, using 92.6 GB. Longest published context on this page is 66k. This page is evidence, not the full buying answer.

Based on 33 external benchmarks; no lab runs yet.

Published runtimes: flash-moe, llama.cpp, MLX, Ollama.

Need a recommendation? Use BuyChecking a target model? Use FitWant a setup answer? Use RunBrowse Macs by exact hardwareNeed the full audit trail? Use Bench

Current published coverage

Published model coverage includes Gemma 4 E2B, Llama 3.1 8B, Gemma 4 E4B, Qwen 3 8B plus 19 more published models. Published runtime coverage on this chip includes flash-moe, llama.cpp, MLX, Ollama. Fastest published row is 158.0 tok/s on Gemma 4 E2B at Q4_K - Medium. Largest published footprint on this page is 92.6 GB for Qwen3-Coder-Next. Longest published context is 66k.

Gemma 4 E2BLlama 3.1 8BGemma 4 E4BQwen 3 8BQwen3-Coder-NextQwen3.5-9B

Other published M5 Max variants: 48 GB · 64 GB

Macs shipping with the M5 Max (128 GB)

MacBook Pro M5 Max 128GB 16-inch

This chip is part of a family. Compare all M5 Max RAM variants →

Raw benchmark rows for M5 Max (128 GB)

Rows sorted by avg tok/s descending. Click source badge to see original measurement page.

ModelQuantRAM req.ContextAvg tok/sPrompt tok/sRuntimeSource
Gemma 4 E2BQ4_K - Medium——158.0 tok/s—MLXref
Llama 3.1 8BQ4_K - Medium——138.0 tok/s—MLXref
Gemma 4 E4BQ4_K - Medium——128.0 tok/s—MLXref
Qwen 3 8BQ4_K - Medium——98.0 tok/s—Ollamaref
Qwen3-Coder-Next8bit87.1 GB4k79.3 tok/s754.9 tok/sMLXref
Qwen3.5-9BQ8_0——78.0 tok/s—MLXref
Qwen3-Coder-Next8bit88.2 GB16k74.3 tok/s1802.1 tok/sMLXref
Qwen3-Coder-Next8bit89.7 GB33k68.6 tok/s1887.2 tok/sMLXref
Qwen3.5-122B-A10B4bit71.9 GB4k65.9 tok/s881.5 tok/sMLXref
Qwen3.5-122B-A10B4bit73.8 GB16k60.6 tok/s1239.7 tok/sMLXref
Qwen3.6-35B-A3BQ4_K - Medium——55.0 tok/s—MLXref
Qwen3.5-122B-A10B4bit76.4 GB33k54.9 tok/s1067.8 tok/sMLXref
Gemma 4 26B-A4BQ4_K - Medium——50.0 tok/s—MLXref
Qwen3-Coder-Next8bit92.6 GB66k48.2 tok/s1432.7 tok/sMLXref
Qwen3.5-35B-A3BQ4_K - Medium——48.0 tok/s—Ollamaref
Mistral Small 4 119BQ4_K - Medium——42.0 tok/s—MLXref
Qwen 3 14BQ8_0——42.0 tok/s—MLXref
Mistral Small 4 119BQ4_K - Medium——38.0 tok/s—Ollamaref
Qwen3.5-27B4bit——31.6 tok/s—MLXref
Qwen 3 32BQ4_K - Medium——28.0 tok/s—Ollamaref
Gemma 4 31BQ4_K - Medium——26.0 tok/s—MLXref
Llama 4 Scout 17B-16EQ4_K - Medium——26.0 tok/s—MLXref
Llama 4 Scout 17B-16EQ4_K - Medium——22.0 tok/s—Ollamaref
Gemma 3 27BQ6_K—8k20.0 tok/s391.0 tok/sllama.cppref
Qwen 3 235B-A22BQ4_K - Medium——18.0 tok/s—MLXref
Qwen3.5-27BQ6_K—8k16.5 tok/s—llama.cppref
Llama 3.3 70BQ4_K - Medium——15.0 tok/s—MLXref
Qwen 3 235B-A22BQ4_K - Medium——15.0 tok/s—Ollamaref
Qwen3.5-397B-A17B4bit——13.0 tok/s—flash-moeref
Llama 3.3 70BQ4_K - Medium——12.0 tok/s—Ollamaref
DeepSeek R1 Distill Llama 70BQ4_K - Medium——11.0 tok/s—Ollamaref
Qwen 2.5 72BQ4_K - Medium——10.0 tok/s—Ollamaref
gpt-oss 120BQ4_K - Medium——7.0 tok/s—Ollamaref

Models tested on this chip

Gemma 4 E2BLlama 3.1 8BGemma 4 E4BQwen 3 8BQwen3-Coder-NextQwen3.5-9BQwen3.5-122B-A10BQwen3.6-35B-A3BGemma 4 26B-A4BQwen3.5-35B-A3BMistral Small 4 119BQwen 3 14BQwen3.5-27BQwen 3 32BGemma 4 31BLlama 4 Scout 17B-16EGemma 3 27BQwen 3 235B-A22BLlama 3.3 70BQwen3.5-397B-A17BDeepSeek R1 Distill Llama 70BQwen 2.5 72Bgpt-oss 120B

Next step

Use Rankings when you need the best overall answer for a Mac, and keep this page open when you need the evidence behind that recommendation.

If you are comparing local Apple Silicon against rented cloud hardware, use AI Data Center Index for current GPU rental context.

Data

benchmarks.json — full dataset  ·  chips.json — chip summaries  ·  benchmarks.csv — CSV export

Data from in-house lab measurements plus community-published benchmarks. See all chip families →