Xiaomi (小米)’s MiMo head Luo Fuli (罗福莉) warns against token price war — “don’t blindly cut prices”

Lead: warning after Anthropic move and MiMo token launch

It has been reported that Xiaomi (小米) Group’s MiMo head Luo Fuli (罗福莉) posted on X yesterday to condemn the recent token (词元) price war sweeping the AI industry. The post followed two industry events: Anthropic reportedly closed a third‑party tool subscription channel for Claude, and MiMo itself launched a new Token Plan. Luo used the moment to argue that rushing to slash token prices and fully opening subscriptions to third‑party agent frameworks risks driving the whole ecosystem into a vicious cycle.

Technical critique: third‑party agents can massively inflate costs

Luo said Claude Code’s subscription design is elegant in resource allocation, but believes it is unlikely to be profitable unless API margins are 10–20x higher. She argued that third‑party frameworks such as OpenClaw materially worsen resource use: they trigger many low‑value, separate API calls carrying very long contexts (often >100k tokens), which multiplies calls per user and destroys cache efficiency. The result, she said, can be API costs “tens of times” the subscription price — not a gap, but a “huge chasm.” In short: cheaper tokens plus open access can look user‑friendly, but can be an economic trap.

Advice and MiMo’s approach: efficiency, not a price race

Luo advised LLM companies not to blindly slash token prices before they understand a loss‑making subscription model. Short‑term, users of these agents will face steep cost pressures that should, in her view, force framework owners to improve context management and cache hit rates rather than rely on subsidized tokens. She also outlined MiMo’s Token Plan: it supports third‑party tools and charges by token — similar in logic to Claude’s extra‑usage packages — but with an emphasis on stable, high‑quality service rather than encouraging impulsive consumption and churn.

Bigger picture: compute limits, geopolitics and ecosystem effects

Luo warned that global compute capacity is struggling to keep pace with agent‑driven token demand. Geopolitical factors — from export controls on advanced chips to supply‑chain tensions affecting datacenter capacity — make an endless race to cheaper tokens unsustainable. She noted that Anthropic’s move, intentional or not, is accelerating a split between open and closed ecosystems, which could force the market toward “more efficient agent frameworks × stronger, more efficient models” rather than purely cheaper token pricing. Is a race to the bottom really in anyone’s long‑term interest? Luo’s answer was a clear no.