Why frontier AI labs are not losing to open source models

On Monday, Jesse Zhang, the CEO of Decagon, published a theory arguing that common assumptions about open source AI in the enterprise are incorrect. Instead of viewing frontier models (high-end, proprietary AI) and open source models (AI models with accessible weights) as direct competitors, Zhang asserted that they represent two phases of the same life cycle. He argued that while more mature AI deployments are switching to lighter models, the overall spend on expensive state-of-the-art models has barely budged. Under this framework, expensive models are used to prove out new use cases, which are then passed to cheaper alternatives as they mature.

Data from cloud platform Vercel and AI model aggregator OpenRouter supports this distinction between token volume and actual token spend. On Vercel’s AI gateway dashboard, data from the past week shows that open source model DeepSeek has taken the lead in token volume, processing just over a third of the tokens on the platform. Over the same period, Z.ai, the lab behind the GLM-5.2 model, reached fourth place. However, Anthropic still accounts for more than half of the overall AI spend on Vercel. Although Anthropic’s share dropped slightly over the past month, it remains dominant. Similarly, on OpenRouter, DeepSeek V4 Flash leads weekly usage with 5.3 trillion tokens weekly, while the frontier model Opus 4.8 processes just over 2 trillion tokens. These figures do not yet capture Nvidia’s Nemotron model, which is also entering the market.

The difference in volume is offset by pricing. OpenRouter data shows that the average token cost for Opus 4.8 is roughly 23x higher than for V4 Flash:

• Opus 4.8 costs $1.37 per million tokens.
• V4 Flash costs just 6 cents per million tokens.

This pricing disparity explains why frontier labs like Anthropic aren’t suffering too much from the rise of open source, at least not yet. While simpler, high-volume workloads shift to cheaper alternatives, frontier labs maintain premium pricing for complex tasks. As Zhang, the CEO of Decagon, put it, “The frontier labs will keep owning discovery. Open source will increasingly own production.” This dynamic allows frontier providers to hold on to the premium token price even as the market bifurcates.

Why it matters

The AI market is bifurcating into a two-tiered system where frontier models maintain premium pricing for complex discovery tasks, while open source models capture high-volume production workloads.