A high-end Mac shows promise for local AI coding models

A Mac with an M3 Ultra chip and 512GB of memory was used to run GLM 5.2 MXFP4 with oMLX and Claude Code. The experience points to local compute as a possible future for AI coding, meaning AI models run on a personal machine instead of only through . such as GLM 5.2 and Hy3 are described as getting closer to leading .

The next big gap may be hardware, not only model quality. The implied benchmark is that this setup feels close to a high-end Claude model such as Opus 4.8.

Key points

  • GLM 5.2 MXFP4 was run on an M3 Ultra Mac with 512GB of memory.
  • The setup used oMLX together with Claude Code.
  • The experience suggests are getting closer to top .
  • Hardware may become the next major limit for local AI coding.
  • The setup is still expensive and unrealistic for many s.
Read original