Python can run Mojo now

Original Article ↗ Hacker News Discussion ↗

Openness, funding, and trust

  • Several commenters like Mojo’s technical direction but are wary of its heavy VC funding and closed-source compiler.
  • Standard library code is Apache 2.0, but the compiler remains proprietary; the company promises open-sourcing in 2026.
  • Some say that’s enough to rule it out for production until then; others note Java, .NET, Swift all started closed and still saw wide deployment.
  • There is concern that VC incentives could lead to lock‑in or pricing shifts once the ecosystem depends on it.

Relationship to Python and the “superset” claim

  • Early marketing framed Mojo as a “Python superset”; current messaging is “pythonic language” / “Python family”.
  • Many wanted to run arbitrary Python and selectively optimize with Mojo, using it as a smooth ramp away from performance bottlenecks.
  • Commenters now see semantics diverging: machine‑sized Int (not Python bigints), missing Python features, and a focus on new systems‑style constructs (ownership, comptime, MLIR).
  • Some employees say full Python compatibility is “deferred” rather than abandoned; others call the superset pitch largely a fundraising/marketing gimmick.

Python–Mojo interop and performance

  • The new feature is effectively: “Python can import Mojo as a C extension,” similar in architecture to Cython and many other tools.
  • Critics argue the headline “Python can run Mojo” is overstated; it’s still compiled code behind the Python C-API.
  • Supporters say the value is smoother tooling: Python‑like syntax, MLIR integration, GPU support, zero‑copy interop with arrays/tensors, and less build/config pain than C++.
  • Microbenchmarks (e.g., factorial) were questioned because CPython’s math functions are highly optimized C with lookup tables.

Mojo vs alternatives (Julia, CUDA Python, Cython, Rust, etc.)

  • Julia is repeatedly cited as already offering:
    • High‑level syntax, strong performance, multi‑vendor GPU backends, and vendor‑agnostic kernels.
  • Others compare Mojo’s role to Cython, Nim+nimpy, PyO3, Triton, Numba, and the growing set of NVIDIA Python DSLs.
  • Defenders argue Mojo’s differentiator is deep MLIR integration and a single language for high‑performance CPU+GPU+accelerator code, aimed especially at AI inference infra.

Target hardware and use cases

  • Mojo currently targets CPUs plus NVIDIA and AMD GPUs (especially datacenter parts), with work ongoing for broader consumer and accelerator coverage.
  • Focus today is inference and AI infrastructure, not general research/training or graphics/games, which disappoints some potential users.

Adoption barriers and community sentiment

  • Enthusiasm: possibility of a Cython replacement, easier GPU programming, non‑CUDA vendor support, and a Python‑like on‑ramp.
  • Skepticism: closed compiler, shifting messaging around Python compatibility, overlap with established ecosystems (Julia, CUDA Python), and reluctance to invest in a VC‑controlled language that might change direction or never fully open.