B.AI tops 1.33T daily tokens, builds ‘AI grid’ settlement layer

B.AI opened no-cost access to a set of advanced models and within days reported daily token throughput exceeding 1.33 trillion. The company said 15-day cumulative volume reached 8.19 trillion tokens and that more than 220,000 new API users joined during the campaign. B.AI reported a total user base of over 2.3 million as of September 3.

The free model lineup included DeepSeek-V4-Flash, DeepSeek-V4-Flash-Vision-Exp, Tencent Hy3, Xiaomi MiMo-V2.5, GLM-5.3-Flash (Ox Alpha) and Qwen3.8-Flash. On September 3, B.AI changed pricing for the two DeepSeek models by adding tiered discounts: a 50% discount during peak hours and an off-peak rate equal to 25% of standard peak pricing. The company kept unlimited free access for GLM-5.3-Flash (Ox Alpha), Qwen3.8-Flash, Tencent Hy3 and Xiaomi MiMo-V2.5. Reported usage remained high after the pricing update.

B.AI uses the term ‘AI grid’ to describe infrastructure it intends to run as a settlement layer for agent applications. The platform aggregates models from multiple providers into a single pool that can be scheduled across workflows. A dual-tier API offers one route with guaranteed direct connections and a lower-cost channel for discounted access. A chat interface provides smart routing that selects the model for each request based on reliability and cost.

Payments and settlement are part of the grid design. B.AI operates parallel Web2 and Web3 payment rails so developers can pay by traditional means or use on-chain methods for verifiable settlements. The platform states developers can choose which payment rail to use for global deployments.

B.AI also integrated the service with developer tools and agent orchestration. The platform is compatible with the Responses API and Codex so a single B.AI API key can invoke both GPT-style models and DeepSeek models in the same workflow. The company described five infrastructure components to support production agents: x402 Payment Protocol for background micro-settlements with a pay-before-response approach; 8004 Identity Protocol for issuing on-chain credentials and logging execution history and credit scores; a Skills Matrix of standardized tool blocks; and two platform assistants, BAIclaw for terminal and multi-agent operations and BAIcode for development tasks from architecture to testing.

B.AI listed Singapore as a base of operations. The company presented the token surge and technical features as elements of its plan to support multi-agent collaboration and large-scale agent deployments.