Apple SiliconLLM Benchmarks

Compare model performance across Apple Silicon chips. Run benchmarks locally, submit results, and explore community data.

Explore BenchmarksGet Started

Fusion Architecture Overview

From hardware to applications, building a complete Apple Silicon AI infrastructure

LiveFuture
Portal
bench.dpdns.org
Web Portal
macOS App
Desktop App
Applications
Fusion-K12 Teacher
Smart Education
Fusion-Finance
Financial Analysis
Fusion-Health
Healthcare
Fusion-Ecommerce
E-commerce Platform
Fusion-Science
Scientific Computing
Ecosystem
Fusion-Plugins Ecosystem
Plugin Ecosystem
Fusion-Bench
Performance Benchmarking
Fusion-Browser
Browser Engine
Fusion-Skills
Skills Library
Fusion-Simulation
Simulation
Fusion-Robot
Robot Control
Fusion-ComfyUI
ComfyUI Workflow
Developer Tools
Fusion-Doc
Smart Documentation
Fusion-Code
Code Generation
Fusion-Design
Design Tools
Fusion-Security
Security Audit
Fusion-cowork
Collaborative Workspace
Fusion-Agent Studio
Agent Studio
Fusion-CLI
CLI Tools
Fusion-Code Modernization
Code Modernization
Core Framework
Fusion-Gateway
Unified API Gateway & Routing
Fusion-Core
Core Runtime
Fusion-Executor
Task Executor
Fusion-Memory
Memory Manager
Fusion-Event
Event Bus
Fusion-Guard
Safety Guard
Fusion-Store
State Store
Fusion-OSAgent
OS Agent
Fusion-MLX
MLX Inference Engine
Fusion-Model Hub
Model Hub
Fusion-Artifacts
Artifacts Engine
Fusion-Multi-Node
Multi-Node Orchestration
Infrastructure
MLX-LM
Language Model Inference
MLX-VLM
Vision-Language Model Inference
MLX
Apple Silicon ML Framework
Metal
Apple GPU Hardware Acceleration

92

Submissions

4

Chip Variants

23

Models Tested

Compare Performance

Side-by-side model comparisons across different Apple Silicon chips, quantization formats, and context lengths.

Community Driven

Submit your benchmark results anonymously. Every contribution helps the community make better hardware and model choices.

Open Source

Built on fusion-mlx, the open-source Apple Silicon inference engine. Fully transparent methodology and data.

Latest Submissions

TypeModelQuantChipMetricValue
SpeedQwen3.8-27B-8bit8bitAppletg_tps6.0 tok/s
SpeedQwen3.8-27B-8bit8bitAppletg_tps6.2 tok/s
Speedmlx-community--Qwen2.5-VL-7B-Instruct-4bit4bitM5 Maxtg_tps54.6 tok/s
SpeedQwen3.8-27B-4bit4bitM5 Maxtg_tps14.9 tok/s
Speedmlx-community--Qwen2.5-Coder-32B-Instruct-4bit4bitM5 Maxtg_tps14.3 tok/s
Speedmlx-community--gemma-4-26b-a4b-it-4bita4bM5 Maxtg_tps43.7 tok/s
Speedmlx-community-Llama-3.2-1B-Instruct-4bit4bitM5 Maxtg_tps203.6 tok/s
SpeedMeta-Llama-3.1-8B-Instruct-4bit4bitM5 Maxtg_tps59.1 tok/s
SpeedQwen3.6-27B-mxfp8mxfp8M5 Maxtg_tps8.3 tok/s
SpeedQwen3.5-9B-4bit4bitM5 Maxtg_tps48.0 tok/s