Launching today
ModelFit
Find the best local LLM your Mac or GPU can actually run
5 followers
Find the best local LLM your Mac or GPU can actually run
5 followers
ModelFit helps you choose the best local AI models for your Apple Silicon Mac, iPhone, or NVIDIA GPU. Get Ollama recommendations optimized for your hardware. This tool has been entirely vibecoded using Claude Code and Opus 4.8 / Fable 5. - Data : 850k Google impressions - ~200 organic clicks/day - CLI just passed 1,000 npm downloads (500 in the last week alone) - All organic, zero ad spend.
So basically I was tired of people asking what is the best llm my mac can run ?
So that's the problem ModelFit is made to solve. To suggest you with the best local llm for your hardware.
I've made it entirely with Claude Code I try to be as accurate as possible but some inaccuracy can still be there and it's an ongoing work because there are new models everyday and it can be out of date pretty quickly.
Let me know what you think about the project, thanks for your time 🙏
How does it actually test the models to make the recommendations, or is it just pulling specs and matching them against known VRAM requirements?
@boranosn4 It's in between, we're not benchmarking each model for each device. That would require some heavy time and investment. Instead the engine does two things : it models how macOS actually allocates unified memory to the GPU and it estimates tok/s from each chip's memory bandwidth adjusted for quantization, MoE active params and KV-cache. The "which model is better" side is pulled from third-party benchmarks (SWE-Bench for example), not my own scoring.