Silicon Score Apple Silicon local LLM rankings
← Canonical rankings

Silicon Score › Chips › M4 (16 GB)

M4 (16 GB) — benchmark record

M4 (16 GB) local LLM benchmarks on Apple Silicon Mac with 14 published rows across 9 models. Peak published speed is 92.0 tok/s on Qwen3.5-4B. Published runtimes include llama.cpp, LM Studio, MLX, Ollama.

14Benchmark rows
9Models tested
92.0Fastest avg tok/s (Qwen3.5-4B)
0Lab benchmarks

Quick take

Best published speed here is 92.0 tok/s on Qwen3.5-4B at Q4_K - Medium. This page is evidence, not the full buying answer.

Based on 14 external benchmarks; no lab runs yet.

Published runtimes: llama.cpp, LM Studio, 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 Qwen3.5-4B, DeepSeek R1 Distill Llama 8B, Llama 3.1 8B, Qwen3.5-9B plus 5 more published models. Published runtime coverage on this chip includes llama.cpp, LM Studio, MLX, Ollama. Fastest published row is 92.0 tok/s on Qwen3.5-4B at Q4_K - Medium.

Qwen3.5-4BDeepSeek R1 Distill Llama 8BLlama 3.1 8BQwen3.5-9BMinistral 3 8BPhi-4 14B

Other published M4 variants: 32 GB

Macs shipping with the M4 (16 GB)

Mac Mini M4 16GBMac Mini M4 24GBMac Mini M4 32GBMacBook Air M4 16GB 13-inchMacBook Air M4 24GB 13-inchMacBook Air M4 32GB 13-inchMacBook Air M4 16GB 15-inchMacBook Air M4 24GB 15-inchMacBook Air M4 32GB 15-inch

This chip is part of a family. Compare all M4 RAM variants →

Raw benchmark rows for M4 (16 GB)

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

ModelQuantRAM req.ContextAvg tok/sPrompt tok/sRuntimeSource
Qwen3.5-4BQ4_K - Medium——92.0 tok/s—Ollamaref
DeepSeek R1 Distill Llama 8BQ4_K - Medium——78.0 tok/s—MLXref
Llama 3.1 8BQ4_K - Medium——75.0 tok/s—Ollamaref
Qwen3.5-9BQ4_K - Medium——72.0 tok/s—LM Studioref
Ministral 3 8BQ4_K - Medium——72.0 tok/s—MLXref
Phi-4 14BQ4_K - Medium——38.0 tok/s—Ollamaref
Qwen3.5-9BQ4_0——4.1 tok/s—llama.cppref
Devstral Small 2 24BQ4_0——3.4 tok/s—llama.cppref
Qwen3.5-9BQ4_K - Small——3.1 tok/s—llama.cppref
Qwen3.5-9BQ6_K——2.2 tok/s—llama.cppref
Qwen3.5-35B-A3BQ4_K - Medium——1.3 tok/s—llama.cppref
Devstral Small 2 24BQ4_1——0.1 tok/s—llama.cppref
Devstral Small 2 24BQ4_K - Medium——0.0 tok/s—llama.cppref
Qwen3.5-27BQ4_K - Medium——0.0 tok/s—llama.cppref

Models tested on this chip

Qwen3.5-4BDeepSeek R1 Distill Llama 8BLlama 3.1 8BQwen3.5-9BMinistral 3 8BPhi-4 14BDevstral Small 2 24BQwen3.5-35B-A3BQwen3.5-27B

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 →