- Published on
No.1 globally! OpenRouter reports DeepSeek-V4-Flash topped 7.22 trillion tokens in calls last week
- Authors
- Name
- DeepHorus
The latest weekly report from OpenRouter, the world's largest AI aggregation platform, shows that DeepSeek-V4-Flash has topped the global rankings with 7.22 trillion tokens of weekly usage, becoming the first Chinese AI model to claim the top spot on the platform's weekly chart. This data was released by OpenRouter's official team on August 4th, quickly drawing attention from the global AI developer community and being regarded as a milestone in the rising international influence of Chinese large models.
OpenRouter is currently the AI API aggregation platform covering the most models globally, integrating more than 400 top closed-source and open-source models, with approximately 8 million registered developers and a total monthly processing volume of around 100 trillion tokens. Previously, Chinese AI models had occupied multiple seats in the top five for 14 consecutive weeks, but none had ever reached the top. This breakthrough by DeepSeek-V4-Flash breaks the long-standing monopoly of American models at the top, further consolidating the collective advantage of the Chinese AI camp on aggregation platforms.
Specific data shows that during the statistical period from July 27 to August 2, DeepSeek-V4-Flash completed 7.22 trillion tokens of usage, far exceeding other models in the same period. The open-source project team OpenCode revealed that the usage of DeepSeek-V4-Flash on the platform surged dramatically, with a single-day processing volume of up to 8 trillion tokens, of which 5 trillion came from free trial quotas and the remaining 3 trillion from paid developer calls. This paid proportion indicates that, beyond research-oriented experimentation, a large number of commercial-grade applications have adopted DeepSeek-V4-Flash as a primary model in production environments.
From a technical perspective, DeepSeek-V4-Flash is a lightweight, high-performance model launched by DeepSeek. Based on a mixture-of-experts architecture, it significantly reduces inference costs through sparse activation mechanisms while maintaining strong code generation and logical reasoning capabilities. Its core advantage lies in cost-effectivenessâon OpenRouter, its pricing is only one-tenth that of leading closed-source models, yet it achieves near or even comparable results across multiple benchmark tests, thus attracting a large number of cost-sensitive startups and independent developers.
The direct beneficiaries of this achievement are the global developer ecosystem. For small and medium-sized teams, the low-cost, high-concurrency characteristics of DeepSeek-V4-Flash make AI-native applications that were previously unaffordable now feasible. OpenRouter's weekly data also confirms this: DeepSeek-V4-Flash's usage increased by over 300% compared to the previous week, while usage for most other models on the platform remained stable or slightly declined, indicating that its new growth mainly comes from incremental demand rather than shifting existing demand.
The industry reacted swiftly. The OpenCode team stated that the explosive growth of DeepSeek-V4-Flash far exceeded expectations, and they are planning to add more inference nodes to cope with the continuously rising load. Tech blogger @Internet Sansan pointed out in an analysis on August 3rd that the top five on OpenRouter are now entirely occupied by Chinese AI models, a trend that has persisted for 14 consecutive weeks, systematically weakening the competitiveness of American models in the open API market. Some domestic developer communities have also celebrated this, viewing it as evidence of Chinese open-source models transitioning from 'chasers' to 'leaders.'
Looking ahead, DeepSeek-V4-Flash's rise to the top may just be the beginning. DeepSeek has confirmed that it will continue to optimize the model architecture and plans to release a larger-parameter version in the short term, further approaching or even surpassing the comprehensive capabilities of top closed-source models. Meanwhile, OpenRouter and other aggregation platforms may adjust free quota allocation rules to balance the ecosystem growth of various models. For global AI application developers, more cost-effective options mean greater freedom for innovation, and the voice of Chinese AI models on the global stage will consequently grow stronger.
From an industry macro perspective, this event reflects deep-seated changes in the competitive landscape of AI large models. Over the past year, American closed-source models dominated the API market, but the rapid iteration of open-source models, particularly those from Chinese teams, is reshaping the supply-demand relationship. The success of DeepSeek-V4-Flash validates the development path of 'cost driving scale, scale feeding capability,' providing a replicable template for other emerging models. As global developers cast their votes through tokens, the balance between technical strength and business logic is becoming the decisive factor in determining victory.
