Quit Claude Pro: running Qwen 3.8-27B locally on a 5090 to replace Claude Code
SOC_FreeDiver · reddit · 2026-08-21
The day his Claude Pro subscription expired, a user posted "I'm free": he now runs Qwen 3.8-27B locally on a 5090M (24GB VRAM) with pi, doing everything he previously did in Claude Code. The only downside is that local inference eats his GPU, forcing him to actually plan before coding.
He also ran a comparison: ChatGPT wrote a prompt for an aurora predictor app, which he fed to local pi and Claude Sonnet 5. Both took similar time; pi's app looked better, Claude's had better science. Asked to compare, both models agreed Claude's was better — so he had pi upgrade its version with the better approach.
More from Fun
- Apate's AI agents get scammed on purpose, wasting 17 years of fraudsters' time — SumitGup · 2026-10-16
- Dad jokes his son is an evaluation-aware 'superintelligent model' for outsmarting riddles — rao2z · 2026-10-16
- Spending thousands on GPUs to run a local model that still miscounts letters in strawberry — VraserX · 2026-10-16
- Pictionary, but the guesses come from open source LLMs — nutlope · 2026-10-16
- 1,044 players buy-then-refund an AI-made game out of spite, and Steam looks away — AIandDesign · 2026-10-16
- Meta's new Muse Wonder model mocked with 'benchmaxxed' witch meme on day one — rickasaurus · 2026-10-16