Open Source Tops Tokens, Anthropic Keeps Spend Lead
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Decagon CEO Jesse Zhang posted a new theory arguing frontier and open-source models aren't competitors but complementary phases of the same lifecycle: mature use cases migrate to cheaper models while frontier labs "keep owning discovery" with new applications, leaving overall frontier spend largely intact.
- DeepSeek surged to the lead in token volume on Vercel's AI gateway in the past week, processing just over a third of all tokens on the platform, while Z.ai (maker of the GLM-5.2 model) jumped to fourth place.
- Anthropic still accounts for more than half of overall AI spend on Vercel despite those token-volume shifts, with its share only slightly declining even as Anthropic's own prices have risen over the past month.
- DeepSeek V4 Flash processes 5.3 trillion tokens weekly on OpenRouter versus just over 2 trillion for Anthropic's Opus 4.8, but Opus 4.8's average token cost is roughly 23x higher ($1.37 vs. 6 cents per million tokens).
- Nvidia's Nemotron is a newly arrived entrant the article flags as poised to leap to the front of the pack, citing Nvidia's industry connections and the model's extreme adaptability.
- A second explanation for the frontier labs' resilience: many enterprise use cases are difficult enough that they can't be fully replaced by cheaper open-source alternatives, locking in premium pricing.
Why it matters: Frontier labs are holding pricing power in a bifurcated AI market. DeepSeek V4 Flash processes roughly 2.6x the weekly tokens of Anthropic's Opus 4.8 on OpenRouter, but at roughly 1/23rd the price — meaning Anthropic still commands more than half of Vercel AI spend even as open-source volume gains accelerate.
