presentofai

DeepSeek V4.1-Flash-1017 hits 33.6T weekly tokens, #1 globally

TL;DR

DeepSeek's V4.1-Flash-1017 hit 33.6 trillion weekly tokens on OpenRouter, an 801 percent week-on-week surge that vaulted it to the global number-one spot and underscored China's widening lead in LLM usage volume.

What happened

  • DeepSeek V4.1-Flash-1017 recorded 33.6 trillion tokens in the week of October 19 to 26, up 801 percent from the prior week, per OpenRouter data cited by nbd.com.cn.
  • The model update was announced October 17, targeting improvements in reasoning, coding, automation, and vision tasks over the September 10 V4.1 Flash base.
  • Chinese models swept all five top spots in individual weekly token rankings: DeepSeek first, MiniMax M3.1 Flash (Free) third at 13.7T, Zhipu AI GLM-5.3 Flash fourth at 13.2T, Tencent Hy4 Preview fifth at 8.97T.
  • Total global LLM usage on OpenRouter reached 188 trillion tokens for the week, up 14.63 percent week on week.
  • Chinese models collectively logged 93.54 trillion tokens versus 22.47 trillion for US models, extending China's lead to 26 consecutive weeks of higher combined volume.

Why it matters

  • A 4-to-1 token volume gap between Chinese and US models on a neutral aggregator platform signals that Chinese LLMs are winning on accessibility and cost, not just benchmarks.
  • MiniMax M3.1 Flash launched October 19 with inference costs down 70 percent and speed nearly tripled, intensifying price pressure on every competitor globally.
  • Chinese models are embedding as critical digital infrastructure in over 40 countries via deployments like the MAZU meteorological AI, creating lock-in that goes beyond consumer chatbot usage.
  • The 801 percent single-week spike shows that model update cadence directly drives adoption: a well-timed release can dominate global usage charts within days.
  • US models grew 29.44 percent week on week but remain a fraction of Chinese volume, suggesting the competitive gap is structural, not cyclical.

What to watch next

  • Whether US frontier labs respond with pricing or speed improvements that close the per-token cost gap driving Chinese model adoption on OpenRouter.
  • Sustained weekly token rankings over the next month: a single spike can be a launch effect, but holding the top position through November would confirm durable demand.
  • Expansion of sovereign or government deployments (like MAZU) into new regions, which would signal Chinese models moving from usage volume to strategic infrastructure lock-in.

Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.