Nanocode: The best Claude Code that $200 can buy in pure JAX on TPUs

What is Nanocode?

A new GitHub discussion for the project "Nanocode" has drawn attention on Hacker News for promising a lightweight, Claude-style implementation written in pure JAX and tuned for TPU execution. It has been reported that the repo — started by salmanmohammadi — presents code that can be run on TPUs and, allegedly, at a cost in the ballpark of $200. The claim is simple and seductive: a compact, reproducible Claude-like stack that developers can run without buying into heavy proprietary tooling.

Why people care

Why does this matter? Because the open-source LLM boom has been all about access and experimentation. Nanocode promises a low-friction way to test Claude-style inference and model behavior on accessible cloud TPU setups. For researchers and hobbyists who want to poke under the hood rather than treat models as black boxes, this is exactly the sort of project that sparks a flurry of forks, benchmarks, and "what if" experiments.

Caveats and community reaction

Of course, caveats apply. It has been reported that performance, safety guards, and parity with Anthropic's Claude are not independently verified in the discussion thread — so take the "best Claude code" phrasing with a grain of salt. Hacker News commenters are doing what they always do: praising the accessibility, questioning claims, and probing for reproducible benchmarks. The key moment here is familiar and electric — open source lowering the bar to entry, and the internet collectively asking, "Can this actually replace my paid API?" Time — and independent tests — will tell.

Sources: github.com/salmanmohammadi, Hacker News