About ModelFit
ModelFit matches local AI models to the hardware you actually own. This page explains who runs it, how the recommendations work, and how the site makes money.
What is ModelFit?
ModelFit is an independent guide to running large language models locally, on Apple Silicon Macs, iPhones, and NVIDIA GPUs. The core of the site is a recommendation engine: pick your device, chip, and RAM, and it ranks open-weight models that genuinely fit your memory budget, with the Ollama command to run each one.
The site is written and maintained by Peter, an independent software developer who runs local models on the same hardware the site covers and answers every newsletter reply personally. Model data is refreshed weekly as new open-weight releases ship. ModelFit is not affiliated with Apple, NVIDIA, AMD, Ollama, or any AI model vendor. All product names and logos are trademarks of their respective owners and are used here for identification only.
How do the recommendations work?
Every device page and the wizard run the same scoring engine over our model database. It weighs how well a model fits your usable RAM (about 70% of total RAM is realistic for inference on smaller machines, closer to 85% at 128GB+), its output quality and speed class, and how established it is in the local AI community. Models that would not fit in memory are penalized hard rather than hidden, so you can see what an upgrade would unlock.
Speed figures are estimates, not measurements. Tokens-per-second numbers on device pages are derived from chip class, memory bandwidth, model size, and quantization, and are labeled as ModelFit estimates wherever they appear. Where we cite a measured number, it comes from a named external source.
Benchmark numbers are source-verified. Any precise benchmark figure we publish (MMLU, SWE-Bench, and similar) must appear in the raw primary source (a model card, an official release post) before it goes live. If we cannot verify a number, we omit it rather than estimate it. Every Ollama command we publish is checked against the Ollama registry before it ships.
Affiliate disclosure
As an Amazon Associate, ModelFit earns from qualifying purchases. Some pages include "Where to Buy" links to Amazon; if you buy through them, we may earn a commission at no extra cost to you. We may also earn a referral commission on cloud-GPU sign-ups (RunPod, Vast.ai). These links are always marked with a disclosure next to them and never influence which models or devices we recommend. Rankings come from the scoring engine described above, which has no knowledge of affiliate revenue.
The site runs no display ads and no sponsored posts. Newsletter sponsorships exist as a separate, clearly-labeled placement (see Sponsor) and never affect rankings or editorial content. See our Privacy Policy and Terms of Service.
Contact
Found a wrong number, a dead Ollama command, or a model we missed? Tell us on X: @modelfit_ or email hi@modelfit.io. Corrections ship fast, usually in the next weekly refresh.
Partnership, sponsorship, or press enquiries: partnerships@modelfit.io.