Magic is working on frontier-scale code models to build a coworker, not just a copilot.

Things we believe

  • Code generation is both a product and a path to AGI
  • AGI safety matters and is solvable
  • To build a great AI product, we need to train our own frontier-scale model
  • Transformers aren’t the final architecture; we have something with a multi-million-token context window

Things we do

  • Stare at PTX and SASS to optimize GPU kernels
  • Keep Kubernetes clusters with thousands of GPUs busy and stable
  • Scour every corner of the internet for data
  • Adapt our sharding framework to train a new architecture
  • Design new hardware-aware algorithms

Why work on ML at Magic?

  • We care a lot
  • You will work on really hard problems
  • We’ve raised $145 million from Nat Friedman, Daniel Gross, CapitalG (Alphabet), and Elad Gil
  • We have thousands of GPUs

If this sounds exciting, come join our team or get on the waitlist!