May 27, 2026 — Xiaomi has slashed the pricing of its MiMo-V2.5 models by up to 99 percent. Its flagship MiMo-V2.5-Pro now costs $0.0036 per million cached-input tokens, $0.435 for cache misses, and $0.87 for outputs. Simultaneously, the company’s 100-trillion-token "MiMo Orbit" developer incentive program, launched on April 28, was fully claimed by May 26, concluding ahead of schedule. While the raw discount grabs headlines, for professional investors and technologists, the true signal lies in the exact numbers Xiaomi landed on.
The Price Table That Dictates the Future
At $0.0036 for cached inputs and $0.87 for outputs, Xiaomi’s MiMo-V2.5-Pro (a 1.02-trillion parameter Mixture-of-Experts model) has achieved near-perfect parity with DeepSeek’s V4-Pro, which recently made its own 75 percent discount permanent at $0.003625 and $0.87, respectively. Furthermore, Xiaomi’s base MiMo-V2.5 exactly mirrors DeepSeek’s V4-Flash economics at $0.0028 for cached input and $0.28 for output.
This deliberate, cent-for-cent price matching establishes a structural floor for "good-enough frontier" performance. Both providers offer one-million-token contexts, OpenAI-compatible APIs, and formidable agentic capabilities—excelling on complex evaluations like Claw-Eval and SWE-bench. When two major competitors independently converge on identical pricing, it signals that premium token margins below the absolute frontier are permanently collapsing.
Motives Driving the Ecosystem Convergence
Xiaomi's aggressive maneuver is driven by four intersecting realities. First, DeepSeek forced the benchmark; with V4-Pro’s pricing permanently reset, Xiaomi had to follow. Second, Xiaomi is prioritizing developer habit formation over near-term API margin. By resetting active Token Plan usages while increasing quotas by up to eight times, Xiaomi ensures its API remains the default sandbox before enterprise procurement hardens.
Third, Xiaomi commands a sprawling consumer and hardware footprint. MiMo functions as a strategic wedge rather than a standalone profit center, allowing Xiaomi to monetize ecosystem optionality across phones, vehicles, and robotics. Finally, the engineering gains are credible. Xiaomi credits the reduction to SGLang HiCache integration and Sliding Window Attention (SWA), drastically reducing data movement across the cache hierarchy—GPU VRAM, CPU RAM, and SSD—to one-seventh of previous levels, while increasing cacheable token capacity fivefold.
Caching and the New Economics of Scarcity
The most critical figure in Xiaomi’s announcement is the sub-cent cached-input cost. This fundamentally alters the economics of repeated-context systems. Applications relying on stable prefixes—such as autonomous coding agents referencing a repository map or support bots using fixed manuals—will see operational costs plummet. Prompt engineering is now inextricably linked to cache engineering; stable prefixes are valuable assets, and cold-context misses are severe inefficiencies.
Yet, as input costs vanish, output tokens and failed execution attempts become the new scarce resources. A cheap model that rambles or requires excessive retries is ultimately more expensive than a pricier, precise alternative. For technical operators, the defining metric is no longer the cost per token, but the cost per completed task—such as dollars per successfully merged pull request.
The Geopolitical Blocker and Investment Conclusion
Despite this structural deflation, Western enterprises remain hesitant. Feedback indicates that the primary blockers are data residency concerns, cybersecurity, and the geopolitical risk associated with Chinese vendors, regardless of open-source MIT licensing. This skepticism creates distinct opportunities for sovereign AI clouds, private inference providers, and compliance-focused routing platforms.
For the investment community, Xiaomi’s parity with DeepSeek forces a fundamental re-evaluation of the AI stack. Inference is rapidly becoming a deflationary commodity input, while high-retention agentic software emerges as the true investable output. The economic center of gravity is moving upward: from model APIs to specialized workflows, and from single-model loyalty to multi-model routing portfolios. Xiaomi’s announcement is not merely a promotional discount; it is a profound market ratification. DeepSeek set the token-price benchmark, and Xiaomi has cemented it.
not investment advice