Today, I'm excited to announce Mixlayer.
Two years ago, I reoriented my life around a singular mission: to build the best inference for open source AI.
We believe that it's incredibly important to have viable, open alternatives to closed-source, proprietary models as we figure out how this technology fits into the fabric of our society. But for open source to remain viable we need more than just great models. We need great platforms around them. The best infrastructure, the best protocols, the best APIs. At Mixlayer we're bringing the rest of this frontier to open source AI models.
Some pieces of software are foundational, they play a critical role in powering almost every product we use today: Operating Systems. Databases. And now: Inference Engines, the code that makes AI models go brrr.
The engine that powers Mixlayer's platform was built from the ground up in Rust. Many have asked: why bother? For us, there was no other option. We knew that to build the platform we wanted it would require innovating at every layer in the stack and questioning every assumption along the way. From the GPU hardware kernels and ML framework all the way up to the gateway powering our API.
Owning the engine allows us to exercise an unusual, almost ridiculous, amount of care and in-depth expertise in how we deliver models to our customers. If you've ever tried to use an open source AI model before, you may have had a bad experience. The quality across the industry is all over the place. Mixlayer's models are built differently and we think our commitment to quality speaks for itself when you use them.
And so now, with the foundation laid, we're ready to take our first public steps. We're launching today with the brand new Qwen 3.5 family of models - from a small but mighty 9B dense model all the way up to a massive, sparse mixture-of-experts 396B model. They work amazingly well for a wide variety of use cases from enterprise RAG to agentic coding and at a fraction of the cost of other AI models.
You can create an account today by going to mixlayer.com.
We could not have reached this milestone without the support of our investors, early customers, and partners:
- Our pre-seed VC Active Capital and Pat Matthews, who's not only backing me for a second time around but has also been an incredible sounding board.
- My former co-founder Nate Stewart and Nathan Booker who have used and tested Mixlayer's models to build their new AI orchestration platform, Runtype.
- And NVIDIA AI, which provided much of the compute for developing our technology. There are many reasons NVIDIA is at the top of their game, one of the big ones is they support their developers like no other company I've ever worked with.
From the bottom of our hearts, thank you.
If you're wondering if open source AI can work for you and want to explore your use-case, I would love to talk to you. Drop me a DM or an email.
Here's to building the rest of the open frontier. 🚀