DEV Community

정상록
정상록

Posted on

whatcani.run — Which AI Models Can You Actually Run on Your Mac?

whatcani.run — Which AI Models Can You Actually Run on Your Mac?

If you've ever tried running AI models locally on your Mac, you know the pain. "Can my M1 Max handle Llama 3.1 70B?" — the answer usually involves scouring Reddit threads, reading GitHub issues, and downloading models to test yourself.

whatcani.run solves this with a simple idea: crowdsourced benchmarks from real users on real hardware.

How It Works

The platform has two parts:

1. Web Interface

Visit whatcani.run, select your device, and instantly see benchmark results from the community. Each model shows:

  • Decode tok/s — actual generation speed
  • Prefill tok/s — prompt processing speed
  • TTFT — time to first token
  • Peak Memory — max RAM usage
  • Score — overall performance grade

Results are labeled intuitively: "Runs great", "Runs well", "Runs ok".

2. CLI Tool

npx whatcanirun
Enter fullscreen mode Exit fullscreen mode

One command. No installation needed. It runs a standardized benchmark (4,096 input + 1,024 output tokens) and lets you submit results to the community database.

llama.cpp vs MLX

The two most popular runtimes for Apple Silicon — and whatcani.run lets you compare them side by side for the same model on the same hardware. Toggle the runtime filter and make data-driven decisions.

Why This Matters in 2026

API costs keep rising. Privacy concerns are growing. Local LLM execution is becoming a practical necessity, not a hobby.

But the biggest barrier to entry? Not knowing what your hardware can handle. whatcani.run removes that barrier with community data.

The Landscape

Tool Focus
whatcani.run Simplest "what can I run?" answer
asiai 7 engines, real-time monitoring
Anubis OSS macOS native, GPU telemetry
Silicon Score Hardware purchase guide

Get Started

# Run benchmark
npx whatcanirun

# Global install
npm install -g whatcanirun
wcir run
Enter fullscreen mode Exit fullscreen mode

Open source on GitHub. Contribute your benchmarks and help the community make better hardware decisions.


What's your experience with local LLM performance on Apple Silicon? Have you found the sweet spot between model size and speed?

Top comments (0)