AI model costs are pushing startups towards cheaper open weigh

Harvey’s gross margin fell from about 50% at the start of this year to minus 50% by June as customer usage of its AI agents spiked, and turned positive again only after it released a model of its own built on Moonshot’s Kimi K3. Abridge, Decagon, Ramp and Rogo are making similar moves, and Sequoia Capital and General Catalyst are funding the shift.

Harvey’s gross margin fell from about 50% at the start of this year to minus 50% by June, Bloomberg reported , citing a person familiar with the matter. The legal startup is valued at $15.6B. Customer usage had spiked after a March update to its AI agents.

In August Harvey released its first in-house model , post-trained on Kimi K3 from the Chinese lab Moonshot. Gross margins turned positive again after that launch and other changes to how it uses AI, people familiar with the work told Bloomberg. Harvey declined to comment on specific financials.

Abridge is building a clinical model on Nvidia’s open weights, and Decagon now routes 80% of customer queries through models of its own. Ramp raised $750M in June and is weighing training for the first time. “ It made absolutely no sense a year ago, ” co-chief executive Karim Atiyeh said. “ It’s starting to make a lot more sense now. ”

At an investor forum in its offices it showed a slide on startups building their own models, with a reminder that Harvey still needs Opus for its hardest tasks. Mistral’s Arthur Mensch spent July arguing that closed models give providers immense leverage , and named Anthropic cutting off Windsurf while building Claude Code.

Legora, in Stockholm, reached $100M in revenue inside 18 months and now serves more than 1,200 firms, and it has never said whose models it runs on. Not everyone thinks the answer matters. “ Having your own model or not is such the wrong question, ” said Menlo Ventures partner Matt Kraning. “ In most cases, it tends to be a lot of cosplay. “