1.33 Trillion Tokens a Day: B.AI Buys Traffic with Free Models, Bets on Being the Settlement Layer of the Agent Economy
Singapore-based B.AI juiced its daily token throughput to 1.33 trillion with a single trick — top-tier models, free — then announced what it actually wants: to become the global settlement layer of the agent economy.
Three Key Facts
According to Business Insider (via a GlobeNewswire release), days after B.AI opened free access to a lineup of top models, its daily token throughput crossed 1.33 trillion. Over a 15-day window cumulative volume reached 8.19 trillion tokens, drawing in more than 220,000 new API users; as of September 3, its total user base had surpassed 2.3 million. The free lineup covers six frontier models: DeepSeek-V4-Flash, DeepSeek-V4-Flash-Vision-Exp, Tencent Hy3, Xiaomi MiMo-V2.5, GLM-5.3-Flash (Ox Alpha), and Qwen3.8-Flash.
Free isn't forever. From September 3, B.AI introduced tiered pricing on the two DeepSeek models: a 50% discount at peak hours, with off-peak rates at just 25% of standard peak pricing; the other four stay zero-cost. The company claims developer momentum didn't waver through the switch, with token throughput still climbing.
B.AI has staked out a bold position for itself: "above all models, below all agents." It wants to abstract models across providers, capabilities and price tiers into one pool of schedulable resources, layered with a dual-tier API structure (official direct routes plus lowest-cost custom channels) to handle orchestration and settlement for agent tasks that constantly switch models. Per CryptoSlate, the platform also integrates TRON and BNB Chain, offering wallet-based payments, programmable identity, and an Agent Wallet that supports asset transfers and token swaps — meaning the settlement layer isn't just API billing; it includes on-chain assets.
WangDou's Take
1.33 trillion tokens a day sounds terrifying until you realize it's all interest paid on one word: free. Burning money for traffic isn't new — what's new is which layer the math is being done on. 220,000 new API users, 2.3 million total; the acquisition cost is just the free DeepSeek and Qwen calls, which are marginal cost for inference providers but acquisition subsidy for B.AI. That spread is the entire bet. The part actually worth watching is those three words: settlement layer. In the agent era a single task hops across multiple models, and developers are still gluing together APIs, bills and rate limits by hand — the equivalent of running between counters with receipts. Whoever unifies multi-model orchestration and settlement into one door becomes the cash register of the agent world. And by shoving an on-chain wallet into it, B.AI is saying it wants a say in the currency inside that register too. Discount the press-release numbers all you like; the direction is right — models are depreciating, pipes are appreciating.
Source: Business Insider / GlobeNewswire