DeepSeek V4.1-Flash-1017 Tops OpenRouter Weekly Call Volume, Up 801% Week-over-Week

Related topics top start Related topics top end Futures promotion begin Futures promotion end Miaopai begin Miaopai end Video player start Video player end Market chart begin Market chart end

Jiemian News reporter | Song Jianan

October 26 news: On the latest weekly global AI large model call volume ranking released by OpenRouter (October 19 to 25), the top spot was taken by the V4.1-Flash-1017 version that DeepSeek updated on October 17. The model's single-week call volume reached 33.6 trillion Tokens, a month-on-month increase of as high as 801%, making it the model with the strongest momentum in the past week and the core driving force behind the overall increment of domestic models.

Public information shows that V4.1-Flash-1017 is based on V4.1 Flash released on September 10, continuing the asymmetric MoE architecture with 552B total parameters, 8B input activation, and 16B output activation, natively supporting multimodality, with a focus on optimizing code, automated Agent, and visual tasks.

OpenRouter weekly call volume ranking

The second place on the list is DeepSeek's older version V4.1 Flash, officially launched on September 10, with weekly call volume of 18 trillion Tokens this week, a month-on-month decline of 45%, indicating that the developer community is rapidly migrating to the new version.

DeepSeek has been making frequent moves recently. First, on October 20 it officially announced the open-sourcing of its self-developed large-scale inference framework DeepInfer, which is precisely the underlying foundation supporting the high-throughput services of the V4.1 Flash series. On October 22, according to a Shanghai Securities News report, DeepSeek plans to raise about 100 billion yuan in its latest funding round, with Tencent and CATL being the largest investors in this round, and the transaction is expected to be officially completed within the next few days.

Third place goes to MiniMax M3.1 Flash, launched on October 19, with weekly call volume of 13.7 trillion Tokens. This model targets complex agents, computer operation, and software development scenarios; the company disclosed that inference speed has increased nearly threefold and call costs have dropped by 70%, mainly aimed at the batch automation task needs of small and medium-sized developers. MiniMax has also recently been accelerating commercialization, with its official brand flagship store launching on e-commerce platforms and tiered subscription packages going online, broadening revenue channels for C-end and small B-end users.

The remaining spots in the top ten list also include Zhipu AI GLM 5.3 Flash, Tencent Hy4 preview, as well as overseas models from OpenAI, Anthropic, and Nvidia.

Zhipu GLM 5.3 Flash was released on August 26, adopting a sparse plus linear attention hybrid architecture with 320B total parameters and 18B activation parameters, focusing on long-context Agent and code review, with weekly call volume of 13.2 trillion Tokens this week, a slight month-on-month increase of 8%. Zhipu's recent GLM 5.3 series has been integrated into Amazon Web Services' Amazon Bedrock, launching a revenue-sharing model with overseas cloud vendors while simultaneously advancing cooperation with multiple domestic cloud platforms.

Tencent Hy4 preview continues to iterate for Agent and office productivity scenarios, maintaining a stable spot on the list. In the overseas camp, the flagship models of OpenAI, Anthropic, and Nvidia still hold the high-end inference market; although their single-week growth rates are impressive, their overall share is less than that of the domestic Flash series.

From the top ten on the list, it can be seen that the Flash lightweight versions have become the absolute main force of platform traffic, and heavy-parameter flagship models are no longer the first choice of developers. In addition, the million-Token long-context window has almost become standard for top new models, with long documents, video understanding, and continuous agent tasks becoming rigid demand. Furthermore, the iteration cycle of large models continues to compress, new versions are densely launched, traffic switching is getting faster and faster, and the advantage window of a popular model keeps shrinking. Competition is concentrated on the contest of effective inference capability per unit cost, and low price and high throughput are the most direct chips for driving developer migration.

Overall, last week's total global AI large model call volume was 188 trillion Tokens, a month-on-month increase of 14.63%. Among them, the weekly call volume of domestic AI large models on the list was 93.54 trillion Tokens, a month-on-month increase of 18.14%; during the same period, the weekly call volume of US models was 22.47 trillion Tokens, a month-on-month increase of 29.44%. The weekly call volume of Chinese large models on this platform has been higher than that of the US for twenty-six consecutive weeks, firmly ranking first.

Looking at the ranking over the past month, changes in the market landscape are even more prominent. A month ago, overseas closed-source models occupied more spots in the front rows of the list, while domestic model spots were relatively scattered. In the past thirty days, domestic vendors have continued to push iterative versions at high frequency, with multiple new Flash products successively breaking into the top ten, gradually squeezing the front-row positions of overseas models.

OpenRouter monthly call volume ranking

Behind the changes are three combined forces driving them. First is architectural innovation: asymmetric MoE, hybrid attention, compressed KV cache and other solutions continue to lower the cost of single inference, turning "adequate capability + low price" into standardized products. Second is the iteration rhythm: domestic vendors push fine-tuned versions on a weekly basis, specifically filling shortcomings in internet access, code, and Agent. Finally, developer-friendly strategies: most new versions are compatible with old APIs and do not raise prices, greatly lowering the migration threshold.

Some analysts mentioned that competition among large models has bid farewell to a pure benchmark-score contest and turned to a contest of effective output per unit Token cost, and that long-tail automation demand is the core source of increment in this round.

It should be noted that the above data only counts public call traffic on OpenRouter's aggregated API channel and does not cover all requests from vendors' own official websites, standalone apps, and private deployments, but it is still an intuitive bellwether for the needs of global independent developers and small and medium-sized enterprises.

English text is being prepared.

Under such industry logic, rises and falls in rankings will become the norm. The version that tops the list today may be surpassed at any time by another rapidly iterating new product. Only by continuously iterating and balancing capability, speed, and cost is the passport for vendors to stay at the table for the long term. news_keyword_pub,stock,sz300750 iframe below main text iframe below main text General statement General statement QR code at end start QR code at end start

QR code at end start QR code at end start Editor name and work code

Editor name and work code