MiMo Director Luo Fuli: Banning third-party tools is inevitable. The Agent era belongs to the most intelligent users of computing power.

According to monitoring by 1M AI News, Xiaomi MiMo’s large model Luo Fuli posted on X analyzing the incident involving Anthropic banning access permissions for third-party tool usage within its subscription bundles. She believes this outcome would happen sooner or later: third-party tools like OpenClaw are extremely wasteful when using Claude subscriptions; a single query triggers multiple rounds of tool calls, with each round independently initiating API requests carrying a complete context of over 100,000 tokens. The actual call volume is several times that of Claude Code itself. Translated into API pricing, the real cost could be dozens of times the subscription price—“not a gap, but a chasm.”

The MiMo Token Plan does not block third-party tools, but charges based on the actual token consumption. Luo Fuli said that opening the door wide to third parties after pushing prices to the lowest level is the trap Anthropic just stepped into. Under pay-as-you-go pricing, third-party tools naturally have an incentive to reduce invalid calls and compress token waste. “This kind of pressure will ultimately turn into engineering standards.”

Her conclusion is that the global compute power growth rate can’t keep up with the token demand created by Agents. The right path isn’t to undercut anyone else on pricing, but to evolve together with “a more efficient Agent framework × a more powerful model.” In the closing paragraph, it says: “The Agent era doesn’t belong to the person who burns the most compute power—it belongs to the person who uses compute power the smartest.”