Using an open model feels surprisingly good
Points and comments are a snapshot, not live.
Running an open model on a private endpoint feels freeing and cuts ties with providers.
The author, a Modal employee, tried Kimi K3 on Modal's new managed endpoints after being dissatisfied with personal Claude/ChatGPT plans. Setting up opencode to point at his own endpoint took five minutes and felt 'like opening vim' after big editors. He describes a surprising feeling of freedom, owning the endpoint, and data traveling directly between his laptop and the endpoint.
He notes he didn't contribute to the feature and wanted to use Kimi K3 instead of upgrading his Claude plan.
What commenters are saying
Many commenters flagged the post as a thin advertisement for Modal, noting the author works there and used company infrastructure. A skeptical camp argues local AI is expensive and unnecessary compared to paid services, while another camp advocates running small open models locally for privacy and control. Several commenters share setups using open models like DeepSeek V4 Flash with custom harnesses for home automation, at lower cost. Cost comparisons and data retention concerns also appear, with OpenRouter's zero-data-retention providers mentioned as an alternative.