Writing code models that run on 128 GB

71 models with published weights that fit in 128 GB — a large Mac or a server card. Counted at four-bit quantisation, weights only, leaving the machine about a third of its memory and a gigabyte for context: room for roughly 136 billion parameters. A mixture of experts is counted in full, because it is held in full even though only a few experts compute.

A large Mac or a server card. This is where the biggest openly published models come within reach — and where the distinction between total and active parameters starts to decide everything, because a mixture of experts computes with a fraction of itself and must still be held whole.

Models and agents for reading a codebase and changing it, rather than writing a snippet. The boards that matter here are the ones that run against real repositories — SWE-bench, SWE-rebench, Terminal-Bench — because a model can write a plausible function and still fail to make a test pass. Note whether what you are looking at is a model or an agent: an agent brings the harness, the file access and the loop, and is priced for it.

Wider

Writing code models Models that run on 128 GB