Sushi Quant Squeezes Qwen3.8 Into 43.8GB, Runs Locally on 64GB Apple Silicon
alexcovo_eth · x · 2026-09-28
Sushi, a custom inference engine for Apple Silicon (a fork of mlx-serve), released its smallest quant yet: Qwen3.8-Flash-Next-Sushi-2.6bpw at just 43.8GB of weights, runnable on 64GB-class Macs.
- Mixes EXL3 and affine quant formats tailored for M5 Max-class chips
- Also offers 3bpw (64GB+) and 4bpw (96GB+) variants
- Open source on GitHub; works standalone or as a guest engine within mlx-serve
More from Infra
- India's Sarvam launches Token Factory, hosting top open models on local infra — HarveenChadha · 2026-11-16
- Gemini 3.5 Flash now runs on air-gapped Google Distributed Cloud, offline up to 12 months — shashib · 2026-11-16
- Gemma.c: full Gemma 1/2/3 multimodal inference in a single C file, ~72 tok/s on CPU at INT8 — This-Purchase-3325 · 2026-11-16
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Jailbroken PS4 runs a 9B open-source model at up to 43 tok/s by talking directly to its 2013-era GPU — AIFlow_ML · 2026-11-16
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16