Built for the people who build MLX.
MLX Network runs on Apple's open-source MLX framework. These programs give the MLX community things it can't get anywhere else: speed data for every Mac, compute, conversions, an API and funding for work already done.
MLX Bench
Open nowThe open leaderboard of how fast every Mac runs every popular model: generation and prompt tokens per second, and peak memory, from M1 to M5. Today anyone can add their Mac with one script. At mainnet, Macs running MLX Node re-run the suite so every number is checked.
Open the leaderboardGrants for past work
Open nowPaid in dollars to the people who already carry the MLX ecosystem: maintainers of libraries like mlx-vlm and mlx-audio, the Swift packages, tools, tutorials and community meetups. We fund work that's already done and useful, judged by outside reviewers, never by stars, pull requests or downloads.
A Builder seat
Open nowA free Core seat (a dedicated seat) for anyone whose code was merged into the MLX repositories before 16 October 2026: ml-explore's MLX, mlx-lm, the Swift and C packages and examples, mlx-vlm, mlx-audio and exo. The cutoff has already passed, so new pull requests can't earn one. One per GitHub account, reviewed by hand.
Cluster time
At mainnetFree credits on pods of Mac Studios linked by Thunderbolt 5, for testing trillion-parameter models, tensor parallelism and distributed fine-tuning with MLX and exo. Very few developers own four Mac Studios; now they won't need to.
Same-day conversions
At mainnetWhen a new open model drops, Macs running MLX Node convert it the same day and measure how much quality each quantization loses against the original, so you can pick a 4-bit build with numbers instead of hope. Published under MLX Network's own Hugging Face account, or contributed to mlx-community only with its admins' agreement.
An API for MLX models
At mainnetAn OpenAI-compatible endpoint serving popular mlx-community models, speech and chat first, run by Macs running MLX Node. Free starter credits for builders, and every request pays the Macs that serve it.
How we work with the MLX community
- We never pay for stars, pull requests, uploads, likes or follows. It breaks GitHub's and Hugging Face's rules and buries maintainers in spam.
- We reward work that already exists, judged by people, and we credit it publicly.
- Everything we measure is open data, with the method published next to it.
- We talk to maintainers before we touch their projects, and we contribute upstream only where they want it.
- MLX Network is independent and not affiliated with Apple Inc. Apple's MLX framework is open source under the MIT license.