# @jmbollenbacher — 2026-09-15

♥5 ↻0 · https://x.com/jmbollenbacher/status/2099949620763697227

GLM-5.3 is 744B params and can be quantized into a usable ~200GB quant.

DeepSeek V4.1 Flash is also 744B parameters, counting the 198B ngram embeddings table. Can it also be quantized into a usable ~200GB quant?

Key question for ds4 as M5 studios arrive @antirez @ivanfioravanti

tags: author:jmbollenbacher, kind:tweet, on:deepseek-v4-1-flash, year:2026
cited on: deepseek-v4-1-flash