I'm not paying $20 for ChatGPT or Claude because a free local LLM does

A writer explains why a free local LLM replaced two $20-per-month cloud subscriptions for everyday tasks. Running Qwen 3.8-27B (Unsloth UD-IQ4_XS build) via llama.cpp on an RTX 4070 Ti Super with 16GB VRAM, the 13.3GB model runs with a 16K context window at roughly 33.7 tokens/sec and handles grammar checks, email summarization, sentiment analysis, parsing scattered numbers into tables, and simple coding tasks (it produced a working Snake game in one prompt). The local setup eliminates cloud limits and costs: no five-hour or weekly caps, no overage charges, and data never leaves the PC. Smaller Qwen and Gemma variants mean options exist for less powerful hardware.

The write-up acknowledges frontier models (Anthropic, Google, OpenAI) still lead on deep reasoning and code-heavy workflows, and that local LLMs don’t top benchmarks, but argues they’re “good enough” for most productivity uses. Trade-offs are hardware and electricity (GPU drawing ~285W) and initial friction to self-host, but those are framed as surmountable compared with recurring subscription fees and privacy concerns. Community comments note occasional failure modes like reasoning loops, but the takeaway is that for many users a properly chosen local model delivers equivalent day-to-day utility at effectively zero software cost.