The AI Market is Bifurcating: Will Open Source Models Replace Anthropic?

While open source models are rapidly evolving in the world of AI, leading laboratories like Anthropic and OpenAI still maintain their dominant position. According to a new theory proposed by Jesse Zhang, head of Decagon, open source models and frontier models are not actually competitors, but complementary systems. This helps to understand the economic balance in the tech world in a completely different way. This is reported by Techcrunch.com reports .

Today, many companies use the most expensive and powerful models when developing their products. However, once the product matures and features are standardized, they prefer to switch to cheaper open source alternatives. Jesse Zhang notes that expensive models are needed to discover new capabilities, while open source models serve to implement ready-made solutions into mass production.

The gap between token volume and revenue

AI gateway data from the Vercel platform partially confirms this theory. In recent weeks, the DeepSeek model has taken the lead in terms of token volume, accounting for one-third of all requests. Also, the GLM-5.2 model from Z.ai laboratory entered the top four. However, in terms of total costs, Anthropic still accounts for more than half of all AI spending on the platform.

OpenRouter statistics show a similar situation. While DeepSeek V4 Flash processes 5.3 trillion tokens per week, the most popular expensive model, Opus 4.8, handled only 2 trillion requests. However, every million tokens of the Opus model cost 23 times more than the V4 Flash model. This means that advanced laboratories are generating more revenue with fewer requests.

NVIDIA and future competition

The market situation is expected to become even more complex with the emergence of NVIDIA's Nemotron model. Thanks to its extensive connections and the flexibility of the model, NVIDIA could soon become a leader among open source models. Nevertheless, companies like Anthropic believe there is no cause for concern yet, as the scope of tasks in the AI field is expanding so rapidly that high-level models are still required for the most complex operations.

In conclusion, the AI economy is becoming a two-tier system. In the words of Jesse Zhang: "Frontier labs lead the discovery phase, while open source models take over production." Although this model currently seems stable, as the intelligence level of cheaper models increases, Anthropic and other giants may have to rethink their strategies.