AI model costs are pushing startups towards cheaper open weigh

Harvey’s gross margin collapsed from about 50% to minus 50% by June after a March update spiked token usage twenty-fold, forcing the company to stop renting external AI models. In August it released an in-house model trained on Moonshot’s Kimi K3, restoring positive margins. Similar shifts are occurring at Abridge, Decagon, Ramp, and Rogo, with Sequoia Capital and General Catalyst backing the move toward cheaper open-weight solutions.