Why the rise of open source AI isn’t hurting Anthropic … yet
TL;DR
TechCrunch frames Decagon CEO Jesse Zhang’s thesis: mature AI deployments often move to lighter open source models, while expensive frontier models are used to test and validate new use cases. The dashboard evidence is mixed but useful: on Vercel, DeepSeek leads by token volume and Z.ai is rising, yet Anthropic still accounts for more than half of AI spend on the platform, according to the article.
Nauti's Take
Anthropic is not automatically losing to open source because the expensive tokens sit at the riskiest part of the workflow: making the first version work. That remains a strong business model as long as customers pay to avoid experimentation pain.
Still, comfort is dangerous. Teams that pour Claude into every pipeline are baking in cost blindness.
The smarter stack uses frontier models for uncertainty and open models for routine.
Briefingshow
For teams, this is a buying strategy, not an ideology fight. Frontier models stay expensive because they reduce friction on new, messy tasks. Once a workflow is stable, cheaper models become attractive.
The real operating skill is model routing: prove the use case, lower the cost, then keep monitoring quality.
Sources