This article is compiled and organized by BlockWeeks.
Market Trends and the Rise of Open Models
The walls may not be closing in on OpenAI and Anthropic as quickly as benchmarks and public opinion suggest.
This article was first published in Galaxy Research's weekly newsletter.
China's open-weight AI models have turned reasoning into a commodity, questioning the sustainability of 60%-80% markups at frontier labs. However, the walls may not be closing in on OpenAI and Anthropic as quickly as benchmarks and public opinion suggest.
Enterprise Preference: Willing to Pay to Avoid Technical Complexity
Decagon co-founder Jesse Zhang recently pointed out that running open models requires technical details that many enterprises would rather pay to avoid. As the old saying goes, "No one ever got fired for buying IBM." Most enterprises could actually build the necessary infrastructure internally, but it is much easier to hand the problem to Anthropic or OpenAI.
OpenRouter Data Shows Divergence Between Usage and Spending
Data from OpenRouter supports this hypothesis. OpenRouter is a single API endpoint that unifies inference service providers for both open-weight and closed-weight models. Since mid-May, the share of open-weight models in token usage has grown by nearly one-third, reaching 75%. However, in terms of estimated dollar spending on the platform, closed-source models still overwhelmingly dominate open-weight models. This indicates that while open models are rapidly gaining in usage, enterprises still tend to choose closed-source solutions in terms of actual costs.
