mlx-coding-bench
Which local LLM runs best for coding on your Mac? The only benchmark measuring both speed and quality for agentic coding on Apple Silicon.
Speed vs Quality
The tradeoff that matters. Top-right is the sweet spot — fast enough for interactive use AND smart enough for real coding tasks.
Rankings
Sorted by weighted quality score (coding, reasoning, tool calling, math — harder problems count more).
| Model | Type | RAM | Quality | tok/s | Coding | Tools | Reasoning |
|---|
Speed Tiers
How fast is fast enough? Color-coded by real-world usability: green = autocomplete-fast, blue = agentic coding, purple = interactive chat, orange = slow.