What happened
Data from AI gateways shows a split in the model market. According to a report from TechCrunch, open-source models like DeepSeek now lead in token volume on platforms like Vercel and OpenRouter. Despite this, frontier models from labs like Anthropic continue to capture the majority of total spend. On Vercel, Anthropic holds over half the spend, while on OpenRouter, its Opus 4.8 model costs roughly 23 times more per token than the volume leader, suggesting it still dominates revenue.
How the room's reading it
The consensus is that this isn't a simple competition. The theory, articulated by Decagon CEO Jesse Zhang, is that open-source and frontier models serve different phases of a product's lifecycle. Frontier models are used for discovery — proving out new, complex use cases where raw capability is paramount. As a use case matures and its requirements are better understood, it can be "graduated" to a cheaper, more specialised open-source model for production scale. This creates a stable two-tiered economy. Frontier labs keep their premium by tackling new problems, while open-source models handle established ones, explaining why top-tier model spend remains high.
Sailfish's take
We think this confirms the right way to build. The debate over 'open-source vs. proprietary' misses the point entirely — it’s not a one-time choice. It’s a continuous process of portfolio management. We've seen teams waste months trying to force an open-source model to work on a novel problem when they should have just used the best tool available to prove the idea had legs. The smart playbook is clear: use expensive, powerful frontier models to find product-market fit for a feature. Once it's validated and running, then you can analyse the task and swap in a cheaper model. If you're not managing a pipeline of features from discovery to production, you're just leaving money on the table.