After two years using Claude and ChatGPT like everyone else, Modal engineer Matthew Saltz had an unexpected experience: running an open model on his own endpoint. And, according to him, it was surprisingly good.

‘It’s liberating, in a way,’ Saltz wrote on his blog. ‘I own the endpoint, and my data goes from my laptop to there and back. It feels like it’s mine. It’s really cool.’

The motivation came from a simple desire: to start a personal project when he got home. Saltz doesn’t have the best Claude or ChatGPT plan for his personal account. At Modal, where he works, the company had just launched Kimi K3 on managed endpoints. Knowing that the Chinese model is considered solid, he decided to test it instead of upgrading his Claude plan.

‘I didn’t directly contribute to this feature, so I hadn’t played with it yet,’ he explained.

In about five minutes, Saltz configured opencode to point to his own endpoint at Modal and was already running. When he started opencode, he felt ‘a clean, pleasant emptiness.’ The best analogy he found was ‘opening vim after spending a long time in a big, feature-filled editor.’

‘It’s as if the tentacles that connected me to other providers had been cut, and I could breathe freely,’ he described, admitting he’s being dramatic, but not exaggerating much.

The feeling was unexpected and motivated him to write about it. ‘It’s weird, lol, and unexpected for me, and that’s why I wanted to write about it.’

Saltz’s account echoes a growing sentiment among developers: the autonomy provided by open models and own infrastructure goes beyond economics—it touches on digital sovereignty. Being able to run a competitive model like Kimi K3 without depending on third-party APIs changes the relationship with technology.

For AIatolah, Saltz’s experience reinforces the importance of the open-source ecosystem and models like the Chinese ones (Kimi, DeepSeek, Qwen) that offer quality without corporate ties. The freedom to choose where and how to run a model is a concrete step towards a more democratic AI.

Kimi K3, developed by Moonshot AI, is one of the Chinese models that have been gaining prominence for its efficiency and openness. Being able to run it on Modal-managed endpoints shows how AI infrastructure is becoming more accessible.

‘Using an open model is surprisingly good,’ concludes Saltz. For those who advocate AI as a common good, the phrase sounds like a mantra.