RSRavid Shwartz Ziv@ziv_ravid7:05 PM · Jul 17, 2026

7/ Bridgewater already showed what this looks like. They fine-tuned an open model on their own financial data with Tinker and got better results than proprietary models at about 1/14 the cost. Very bullish on Thinking Machines. Next time: the neolab jungle 🥳

RSRavid Shwartz Ziv@ziv_ravid7:05 PM · Jul 17, 2026

6/ I think the Thinking Machines playbook is the strongest. They already have customers on Tinker; the open model makes those services better, and nobody knows how to fine-tune this model better than the people who trained it.

RSRavid Shwartz Ziv@ziv_ravid7:05 PM · Jul 17, 2026

664TECH

4/ With the Chinese labs, we honestly don't know why. Winning the AI race against the US? A side project because it's cool? Attracting local talent? Meanwhile, in the West, Hugging Face made a big mistake staying out of the open model race, and Mistral just doesn't have the money for frontier models.

RSRavid Shwartz Ziv@ziv_ravid7:05 PM · Jul 17, 2026

664TECH

3/ But the level of openness is only half the story. The other half is who releases and why. Big companies like NVIDIA and Microsoft mostly do it for PR and to get people playing with their tech. Usually, it's a decision by a few researchers inside the company, and it can change next week.