Frontier AI.
Yours to run.
Meld Labs is an AI research lab building infrastructure for inference, RL post-training, agent sandboxes, and training. Run it on hardware you control — from a single MacBook to GPU servers in your company’s private cloud. Your models and data stay yours. Our managed cloud is next.
A 125B model.
On a MacBook.
Meld Turbo runs Qwen3.8-Flash-Next — a Mixture-of-Experts model with 125B parameters, 6B of them active per token — on a single Mac, behind an OpenAI-compatible API. No cloud, no API key, nothing leaves your Mac.
- Parameters
- 125B
- Tokens / second
- ~45
- Context
- 128K
Four directions.
Starting in the open.
Model inference
Take models from experimentation to production. Build around your workload, not around infrastructure constraints.
Learn moreRL post-training
A guided path from model and reward configuration to reinforcement learning, evaluation, and export.
Learn moreAgent sandboxes
Give agents an isolated environment to execute code, work with tools, and turn decisions into actions.
Learn moreTraining & research
Explore compute-intensive training and research workflows with an infrastructure layer designed for AI.
Learn moreResearch worth sharing.
Where the time goes
on an M2 Max.
Why dense K-quants stall at small batches, how a smaller draft vocabulary pays off, and everything we tried that didn’t help.
Read the reportBring your big idea.
Investors, collaborators, and builders — we’d love to hear from you.
Get in touch