145 days into the trillion-level call scale, the growth speed of B.AI has attracted industry attention.

Written by: B.AI

B.AI Steps into the Trillion-Level Token Era, Establishing the Foundation for the "Intelligent Global Settlement Layer"

Recently, the AI infrastructure platform B.AI has seen its daily token throughput surpass 1.1 trillion for the first time. This not only marks an important growth milestone since the platform launched nearly five months ago, but also signifies that B.AI has begun to enter the ranks of trillion-level AI calling infrastructure. As global developers continue to unleash demand for multi-model calling and Agent applications, B.AI is expanding from a unified model entry towards underlying aspects such as model routing, call metering, and payment settlement. Recently, B.AI further clarified its product positioning as an "intelligent global settlement layer," which undertakes basic services such as model routing, call metering, and payment settlement between models and Agents.

Currently, B.AI's user base has surpassed 2.28 million. Since its launch on April 9, the platform's daily token processing scale has grown from 21.1 million to over 1.1 trillion, achieving a leap from the millions to trillions in just 145 days. In comparison, the global leading multi-model service platform OpenRouter took about 2 years and 9 months to reach a daily trillion token scale from its launch. Using this public timeline as a reference, B.AI has reached the same throughput level at approximately 7 times the speed of OpenRouter.

From Millions to Trillions, B.AI Enters the Scaling Stage of AI Infrastructure

For a model service platform, token throughput better reflects actual usage intensity than merely the number of registered users. Especially in code generation, long text processing, and Agent workflows, a single task may continuously call models, tools, and context. Whether the platform can stably handle high-frequency, high-concurrency requests directly determines its capability to progress towards the infrastructure layer.

The recent growth of B.AI illustrates this characteristic. From August 17 to August 31 alone, the platform's daily token throughput increased by about 242 times, with a total token processing volume reaching 6.86 trillion and over 196,000 new API users added. Compared to the initial launch phase, where traffic primarily came from chat and single model calls, an increasing amount of traffic now originates from API and ongoing developer tasks and Agent workflows.

The rapid expansion of model supply is also a significant driving force for growth. Recently, B.AI has continuously opened domestic models such as DeepSeek, GLM, Qwen, Kimi, MiniMax, Tencent Hunyuan, and Xiaomi MiMo, offering free calls for some popular models. Developers can test various models under a unified API system, switching or combining based on reasoning capability, context length, response speed, and cost, without needing to maintain separate accounts and keys for different vendors.

This also makes the comparison between B.AI and OpenRouter more meaningful in reality. Both prioritize multi-model access and unified APIs, but B.AI is further extending its capabilities into payment, settlement, and Agent infrastructure. The recent breakthrough of daily token throughput exceeding one trillion first validates the model calling and traffic bearing capabilities, providing support for its evolution towards deeper AI infrastructure.

From Model Entry to Global Settlement Layer, B.AI Extends Its Capabilities to the Agent Economy

If unified model calling addresses the question of "what AI uses," then after Agents enter large-scale application, they also need to solve issues like "how to call, how to pay, how to settle, and how to establish identity." As the calling scale continues to expand, B.AI's previous product layout surrounding model routing, metering, payment settlement, and Agent infrastructure has begun to reveal clearer scale effects.

B.AI is also further expanding its infrastructure capabilities, extending from unified model access to model routing, call metering, and payment settlement, summarizing this product direction as the "intelligent global settlement layer." The underlying logic is that in the future, an Agent completing a task may need to continuously call different models, data services, on-chain tools, and even other Agents, with each call corresponding to resource consumption and value exchange.

Around this direction, B.AI has already built a two-tier capability for unified API and payment settlements. On one hand, developers can call different models using a single account and API; on the other hand, the platform supports traditional payment methods such as bank cards, WeChat, Alipay, and UnionPay, as well as multiple blockchain networks like TRON, BNB Chain, Ethereum, and various crypto asset top-ups. For individual developers and teams in different regions, model calling and payment can be completed within the same platform.

On the Agent side, B.AI is also continuing to advance capabilities such as x402 payment, on-chain identity, Agent Wallet, as well as MCP, Skills, enabling Agents to not only call models and tools in the future but also to independently complete payments, settlements, and on-chain operations. Compared to simply providing a list of models, this system is closer to a "basic service layer" generated around AI calls.

From 2.28 million users to daily processing volumes exceeding 1.1 trillion tokens, B.AI has already accumulated a considerable real call amount. As the call scale continues to expand, the market's observation metrics for such platforms are also changing. In addition to the number of models and token throughput, API stability, model scheduling, payment settlement, and Agent service capabilities are becoming important indicators for assessing platform infrastructure capabilities.