A San Francisco company with a few dozen employees switched 100% of its business running on Claude to DeepSeek v4, as documented in a screenshot posted by CEO Flo Crivello on X. The decision was driven by the company's inference costs being higher than its total salary costs, with Crivello publicly stating in April that the company's inference costs were already higher than the salary costs. The team evaluated for six to nine months, considering among others Kimi K2.5 and GLM-5.1, and finally decided on DeepSeek v4. The workload for the migration was "100x more work than we thought," involving numerous online and offline evaluations, gradual introduction into the gray area, and prompt adjustments. In Lindy's core applications such as email classification and pre-writing responses depending on the user's mood, DeepSeek performed beyond expectations, though in complex automation processes, Claude is still stronger. After the switch, the cost curve "fell like a cliff." This incident happened almost a month ago, and its effect became noticeable at the end of June, when DeepSeek open-sourced the DSpark speculative decoding framework, increasing generation speed by 60% to 85%, followed by the update of Wayfinder Router.