Open model use feels surprisingly liberating, dev writes
Original: Using an open model feels surprisingly good
Why This Matters
User sentiment around data ownership and open inference endpoints is becoming a real adoption driver for open models.
Developer Matthew Saltz, a Modal employee, shares that switching from Claude and ChatGPT to running Kimi K3 on his own Modal managed inference endpoint via opencode took under 5 minutes and felt unexpectedly freeing, with data flowing only between his laptop and his own endpoint.
Matthew Saltz, who works at Modal, wrote on July 27, 2026 that after two years of using Claude and ChatGPT, he decided to try Modal's newly launched Kimi K3 managed endpoint instead of upgrading his personal Claude plan. Using the open-source coding tool opencode, he pointed it at his own Modal endpoint in roughly 5 minutes. The experience felt unexpectedly liberating — he described it as owning the data pipeline end-to-end, with information traveling only from his laptop to his endpoint and back. He compared the sensation to opening vim after spending time in a large, feature-heavy editor, or to having ties to major AI providers cut so he could 'breathe freely.' Saltz notes he did not directly work on the Kimi K3 launch feature at Modal. He describes himself as not a traditional open-source advocate, making the emotional reaction surprising even to himself. Modal launched Kimi K3 on managed endpoints the same day he wrote the post.