Recently, B.AI, a next-generation AI infrastructure platform, launched its "Large Model Computing Power for All" initiative, sparking an unprecedented surge in large model usage among developers worldwide. Within just a few days, B.AI’s total daily token throughput历史性突破 1.33 万亿大关, reaching over 10.86 million API calls in a single day. Over the past 15 days, B.AI’s core throughput experienced explosive growth of 289 times.
Achieving this milestone marks the full launch of B.AI’s “Computing Power for All” initiative, but this is only the first step in B.AI’s strategic roadmap. B.AI’s ultimate vision is to transcend the role of a mere computing power distribution channel and establish a core foundational layer—the Global Intelligent Settlement Layer—enabling seamless cross-node collaboration, orchestration, and value distribution for vast numbers of agents in complex business environments.
As the AI industry fully enters the era of high-frequency collaboration among agents, and as concurrent demands grow increasingly massive, major foundational large models resemble isolated "power plants" scattered across the landscape—while B.AI becomes the "AI grid" connecting this entire era. With forward-looking vision, B.AI precisely positions itself “above all models, below all agents.” By deeply integrating diverse top-tier models and full-stack components, B.AI establishes the most robust and irreplaceable infrastructure foundation for the large-scale deployment of agents and the explosion of productivity.
Daily throughput exceeds 1.33 trillion! B.AI's "Computing Power for All" campaign sparks a global surge in usage.
Recently, B.AI's initiative to offer free access to its large models has rapidly gained traction within the developer community, becoming the most talked-about topic in the industry today. This unprecedented, democratizing “compute狂欢” campaign has not only driven exceptional platform engagement but also set a series of striking, record-breaking milestones.
Within just a few days, the number of API calls on the B.AI platform surged dramatically, driving the platform’s daily token throughput to a strong breakthrough of 1.33 trillion, with a cumulative token throughput of 8.19 trillion by the 15th and over 220,000 new API users added. As of September 3, total platform users on B.AI exceeded 2.3 million. This traffic surge, ignited by “compute for all,” is setting new benchmarks for AI infrastructure adoption through the most authentic market response.
This massive scale of usage is directly attributable to B.AI’s highly thoughtful and precisely targeted free-tier matrix that addresses developers’ cost concerns. During this event, B.AI fully opened access to its lineup of six cutting-edge, top-tier models, offering users unlimited, zero-barrier access.
This production-grade matrix precisely addresses the full spectrum of advanced application requirements: GLM-5.3-Flash (Ox Alpha) and DeepSeek-V4-Flash-Vision-Exp provide powerful support for multimodal vision and ultra-long text parsing; DeepSeek-V4-Flash and Qwen3.8-Flash together deliver high-concurrency response and complex code logic generation; complemented by Tencent Hy3, optimized for Chinese-language foundations, and Xiaomi MiMo-V2.5, designed for universal scenario adaptation. From ultra-fast text and code reasoning to advanced visual parsing, B.AI’s premier lineup completely removes cost barriers for developers aiming to build complex AI applications.
It is precisely this barrier-free, top-tier model matrix that has directly fueled the surge in developer adoption of B.AI. In the past, high token costs severely limited the deployment of complex agent architectures. Now that B.AI has completely eliminated computational costs, developers can finally deploy agent architectures capable of long-chain tasks—such as autonomous planning, tool invocation, and reflection and correction—on a large scale in real-world production and testing environments.
Notably, starting September 3, B.AI adjusted the pricing for the DeepSeek-V4-Flash and DeepSeek-V4-Flash-Vision-Exp models, introducing a tiered discount system (50% off during peak hours and as low as 25% of the official peak rate during off-peak hours), while maintaining free access for four other large models: GLM-5.3-Flash (Ox Alpha), Qwen3.8-Flash, Hy3, and MiMo-V2.5. Despite this transition to a commercial model, developer usage remains robust, with the platform’s total token throughput continuing to grow steadily.
This fully demonstrates that this wildly popular campaign is not only a concentrated release of computing power红利, but also an excellent “barometer” of industry infrastructure evolution. With a single-day volume of 1.33 trillion tokens, it clearly shows that AI applications have officially moved beyond the simple “Q&A era,” and B.AI is now fully empowering the large-scale deployment of advanced AI Agents through its solid technical foundation and refined service mechanisms.
B.AI builds an "AI power grid" to solidify the global settlement layer for the agent economy.
B.AI’s “Computing for All” is just the beginning, not the end. For B.AI, this “computing frenzy” is also an extreme stress test of the platform’s foundational capabilities. As countless developers flood the platform simultaneously, and long-chain agent tasks flow through it at unprecedented density, the true challenge emerges: How can this surge be transformed from a one-time event into a sustainable, evolvable, and measurable systemic capability? At this pivotal moment, B.AI has chosen to look deeper, committing to build a full-stack infrastructure capable of powering the agent era and solidifying its role as the central hub of a global computational settlement layer.
Traditionally, the role of infrastructure has been to ensure “stable supply.” But in the Agent era, the very meaning of infrastructure is undergoing a qualitative transformation. A typical agent task often requires frequent switching between multiple models: using a vision model to interpret inputs, a reasoning model to plan steps, a code model to generate executable logic, and a lightweight model to produce formatted outputs. This “multi-model relay” workflow means that no single model provider can independently support an entire agent pipeline. Developers are forced to confront the fragmented reality of multiple API protocols, billing systems, and rate-limiting rules.
The settlement layer built by B.AI precisely bridges this gap. With a forward-looking perspective, B.AI positions itself precisely "above all models and below all agents," abstracting models from different vendors, with varying capabilities and cost structures, into a single schedulable resource.
Leveraging a tiered API system based on “official stability assurance, self-selected lowest pricing,” developers can enjoy the reliability of direct official connections while flexibly selecting high-quality offerings with varying discount tiers from a rich model lineup, achieving the same task quality at more competitive costs. Paired with the Chat interface’s “intelligent routing” feature, B.AI has built a full-stack “AI power grid,” ensuring that every agent task request is precisely routed to the most suitable model with reasonable cost and stable performance.
The settlement capability of this grid equally accommodates both Web2 and Web3 logic. On the Web2 side, developers can leverage familiar traditional payment methods to easily complete deposits and billing, seamlessly integrating into existing business workflows; on the Web3 side, B.AI provides global developers with decentralized, verifiable, and low-friction payment options through on-chain payment channels.
The parallel integration of two payment systems enables developers and agent applications with different backgrounds and infrastructure preferences to find the most suitable settlement pathway on this network, truly achieving “one integration, global circulation.”
Strike at the core of productivity: B.AI’s full-stack infrastructure redefines the Agent collaboration network.
After establishing a robust computational routing and global settlement foundation, B.AI has successfully transitioned from infrastructure deployment to productivity transformation, fully empowering agents to achieve efficient collaboration across diverse business scenarios and completing the most critical execution-layer puzzle for the full-scale explosion of the agent economy.
The core of this puzzle is support for Codex integration. As the central tool for code execution and agent action, Codex holds definitive technical advantages in understanding project architecture, proactively planning, and executing complex development tasks. Currently, the B.AI API is compatible with the Responses API. Developers can use their B.AI API key within Codex to access B.AI’s supported GPT series models, as well as the highly popular DeepSeek series models.
Through this seamless integration, the technical team can directly embed these two leading code-powered models into their daily coding workflows, meeting multifaceted needs such as code generation, logic comprehension, bug debugging, system refactoring, and development collaboration—all in one place—delivering developers a complete end-to-end experience from model selection to code deployment.
Seamlessly connecting computing power with the application layer is only the first step in unlocking the productivity of large models. Beyond this, to realize the agent economy, it is essential to overcome the barriers to multi-agent collaboration under task chains. To enable vast numbers of agents to operate stably in real-world production environments, B.AI has built a full-stack infrastructure comprising five core components, equipping agents with an extremely precise operational system:
- x402 Payment Protocol: Enable automated micropayments between agents. Built on the HTTP 402 standard, this protocol innovatively introduces a "pay-before-response" mechanism, allowing seamless, high-frequency on-chain micropayments and settlements in the background when making API calls or collaborating on computational resources across agents.
- 8004 Identity Authentication Protocol: Mint a unique "digital ID" for each Agent. The 8004 protocol generates objective, verifiable identity credentials for each Agent on-chain, recording their usage history and credit score, thereby completing the critical credit loop for cross-node collaboration in the smart economy.
- Skills "Skill Blocks": Equip agents with out-of-the-box toolchains that deeply integrate with the MCP Server. A standardized Skills matrix enables agents with highly modular tool invocation capabilities, fully bridging the final mile from "thought and decision-making" to "action execution."
- BAIclaw and BAIcode: Native super assistants of the platform, BAIclaw and BAIcode represent the ultimate expression of B.AI’s underlying architecture and orchestration capabilities, designed as super-apps for end users and developers. BAIclaw focuses on all-in-one intelligent terminals and multi-agent collaboration, deeply integrating a leading large model matrix; users simply input natural language commands to automatically trigger Agents that execute complex on-chain economic activities and provide 24/7 custody. BAIcode, as an advanced AI programming tool, leverages the top-tier intelligence of large models to deliver a full-cycle闭环—from requirement understanding and architecture planning to code generation and testing.
From the surge in token throughput triggered by the free-order狂欢 to the establishment of its architecture as a "smart global settlement layer"; from seamless Codex integration to the full-stack infrastructure built by x402, 8004, Skills, and native assistants, B.AI demonstrates not only a robust technical solution but also a clear vision of the future. In this industry transformation driven by AI, B.AI consistently positions itself "above all models, below all agents," transforming fragmented and dispersed computing power into a system-level network capability that is highly available, transferable, and settleable.
When millions of agents flow seamlessly and collaborate freely across this infrastructure, with millisecond settlement, AI is rapidly moving beyond its role as a one-way responsive tool and truly evolving into an economic agent driving a leap in social productivity. B.AI is not only solidifying the present of accessible computing power but also laying the ultimate foundation for the large-scale explosion of the agent economy. The future is here—and this is just the beginning.