Ask HN: replacing cloud models with a local model for daily coding

Category: Pulse
Status: discussion
Sources: HN discussion
Summary: A 1,053-point Ask HN thread asks whether anyone has replaced Claude or GPT with a locally run model for everyday coding. The consensus in the comments is "not yet for interactive use."
Comments: Commenters reported local models still trail cloud offerings: Gemma 4 on an Apple M4 ran at much lower tokens-per-second than cloud APIs; one user runs DeepSeek V4 Flash on dual RTX Pro 6000 Blackwell GPUs at about 160 tok/s but uses it for auto-write-then-auto-review rather than interactive editing; harder tasks (an AVX-512 bit-matrix transpose) that cloud models solve easily tripped up local Kimi 2.6 and GLM 5.1. Several pointed to attaching OpenRouter to an existing agent to test open-weight models without local hardware, and cited gaps in enterprise tooling for selecting and running local models.