Millie's ternary 35B MoE hits 56% on SWE-bench Verified, runs local coding agent on iPhone 17 Pro

MannyKayy · x · 2026-09-10

LLMs For All released Millie: a complete local coding system (models + runtime + CLI) for macOS and Linux. Its ternary MoE 35B model, derived from Agents-A1 on the Qwen3.5-35B-A3B architecture with 3B active parameters per token, solves 56% of SWE-bench Verified and runs at 261 input / 22 output tokens/s on an iPhone 17 Pro. Millie CLI reads/edits files, runs commands, and works through tasks autonomously; it needs a 16GB+ RAM Mac or a PC with 4GB+ VRAM.

Original post →

More from coding & agent

coding & agent channel →