July 27, 2026
Open-source, open wounds
Using an open model feels surprisingly good
He ditched the big AI apps and commenters instantly turned it into a freedom fight
TLDR: A developer says using an open AI system on a setup he controlled felt surprisingly liberating compared with paying for the big-name apps. Commenters turned that feeling into a full-on debate about freedom, rented hardware, and whether cheaper open tools are finally good enough to threaten the giants.
A developer tried running an open AI model through his own setup and came away sounding like someone who’d left a noisy city for a quiet cabin. In his blog post, he says using a model he could point to himself felt "freeing"—less like borrowing someone else’s shiny tool, more like having a space that was actually his. The practical reason was simple: he didn’t want to pay for a pricier personal subscription, so he tested an open option instead and loved the feeling almost immediately.
But the comments? That’s where the fireworks started. One camp basically yelled, "Finally!" saying it’s wild that a crowd once obsessed with building everything themselves took so long to get excited about self-run AI. Another camp hit back with the killer question: is it really "yours" if it still runs on rented hardware? That turned a cozy personal essay into a mini identity crisis about ownership, control, and whether leasing a machine is just subscription culture wearing a fake mustache.
Then came the performance nerds, tossing around model names like sports fans arguing quarterbacks. Some insisted open models are now close enough to the big paid ones for everyday coding, while others said the expensive leaders still win when the task gets messy. The funniest vibe in the thread was how quickly a simple "this feels nice" post became a referendum on tech values, old-school hacker pride, and who’s secretly posting an ad.
Key Points
- •The author says they have used Claude and ChatGPT for about two years before trying an open-model workflow.
- •The article states that the author works at Modal, which launched Kimi K3 on managed endpoints that day.
- •The author chose to try Kimi K3 for a side project instead of upgrading a personal Claude or ChatGPT subscription.
- •According to the post, opencode was connected to the author’s own Modal inference endpoint in about five minutes.
- •The author describes the resulting experience as feeling more controlled and personal because the endpoint was their own and data moved only between their laptop and that endpoint.