Open-source AI growth does not yet displace frontier labs like Anthropic, data from Vercel and OpenRouter suggest

A Decagon CEO’s post argues that frontier labs and open-source models are not competitors but sequential phases in an AI lifecycle: expensive frontier models are used to prove new use cases, which are then handed off to cheaper open-source alternatives as they mature. The author contends that frontier labs will continue to "own discovery" while open source increasingly "owns production."

Dashboard data from Vercel’s AI gateway indicates that DeepSeek has recently surged into the lead for token volumes, now processing just over a third of tokens passing through the platform. Over the same period, Z.ai’s GLM-5.2 model ranked fourth in token volume. Despite this shift, Anthropic still accounts for more than half of overall AI spend on Vercel’s platform. Anthropic’s share has declined slightly over the past month, but the drop is not significant.

OpenRouter’s dashboard, which covers a broader market segment, shows DeepSeek V4 Flash processing 5.3 trillion tokens weekly, compared to 2 trillion tokens for Anthropic’s Opus 4.8. OpenRouter does not rank models by total spend, but it reports that the average token cost for Opus 4.8 is roughly 23 times higher than for V4 Flash ($1.37 per million tokens versus $0.06 per million tokens). This suggests that Opus 4.8 likely still captures the majority of spending despite lower token volumes.

The article notes that Nvidia’s Nemotron model may soon lead in usage due to Nvidia’s ecosystem advantages and the model’s adaptability, though its impact on spending shares is not quantified in the available data.

The observed pattern supports a two-tiered AI economy: frontier labs maintain premium pricing and early-stage dominance, while open-source models capture mature, cost-sensitive production workloads. Analysts cited in the piece suggest this structure could persist, with frontier providers retaining the most desirable part of the marketplace—premium token pricing—at least in the near term.