Why the rise of open source AI isn’t hurting Anthropic … yet
TL;DR
The core idea: frontier models such as Anthropic keep owning the expensive discovery phase, while open or cheaper models absorb mature production workloads. Vercel AI Gateway data from July 8 shows the split: DeepSeek leads token volume with 34.8%, while Anthropic still captures 57.1% of spend. OpenRouter points in the same direction: DeepSeek V4 Flash handles far more tokens than Opus 4.8, but Opus remains much more expensive per million tokens.
Nauti's Take
This is not open models beating Anthropic. It is a routing problem.
Serious AI teams should sort workflows by risk, quality needs, and cost, not by provider loyalty. Anthropic is in a comfortable position, but not a permanent one: if open models move faster into harder tasks, the premium zone gets smaller.
The lead is real, but rented, not owned.
Briefingshow
For teams, the model choice becomes less ideological and more operational. You test new, uncertain tasks with expensive top models, then move stable routines to cheaper models once the workflow is understood. Anthropic does not automatically lose revenue as long as new use cases keep entering the premium layer.
Sources