Xiaomi MiMo v2 Pro: trillion-parameter AI model that outperforms on coding
Xiaomi’s MiMo-V2-Pro (and its Omni sibling) was quietly released on March 18, 2026, targeting the “agent era.” The model briefly went viral on OpenRouter under the anonymous name “Hunter Alpha,” which many users speculated was DeepSeek V4.
Xiaomi’s Luo Fuli (head of MiMo) later clarified that Hunter Alpha was an early internal test build of MiMo-V2-Pro. Xiaomi’s stock reportedly jumped about 5.8% after the reveal.
MiMo-V2-Pro is described as having 1T+ total parameters with 42B active per request via mixture-of-experts, a context window up to 1 million tokens, and a multi-token prediction layer to speed generation. It is closed-source, though Xiaomi left open the possibility of future release.
Benchmark highlights cited in the review: SWE-bench Verified ~78% (vs Claude Opus ~80.8%), ClawEval ~61.5 (approaching Opus ~66.3), and PinchBench ~81.0 (near the top globally). Pricing cited: ~$1 per million input tokens and ~$3 per million output tokens (up to 256K context).
In hands-on tests, Xiaomi MiMo v2 Pro produced a long, richly detailed creative story in one shot and generated a working “stealth game” on the first attempt, including sound/MIDI without losing coherence. The review also notes occasional math issues and inefficiency at times.
Separately, Xiaomi partnered with Sei to preinstall crypto wallets on devices across Europe, Latin America and Southeast Asia—an adoption angle, but not directly tied to MiMo-V2-Pro’s model performance.
Neutral
这则新闻本质上是 AI 模型产品评测与发布信息，短期对加密市场的直接“币价驱动”有限；更偏向科技板块情绪与生态叙事。
1) 与市场的直接关联较弱：MiMo-V2-Pro 的参数规模、上下文长度与 SWE-bench/代理类基准数据，主要影响的是 AI 产业竞争格局，并不等同于某个主流链/代币的立即增量需求。
2) 但存在“温和正面”接触点：文中提到小米与 Sei 合作在设备预装钱包，这是可落地的用户触点，有利于长期采用叙事；不过新闻未给出量化落地规模（例如装机量、活跃钱包数、资金流入），因此难以转化为可交易的短期利多。
3) 类似历史经验：过去当大型科技公司“宣布/发布”AI能力或与链生态合作时，通常会带来阶段性叙事溢价，但在缺少明确链上指标（TVL、手续费、用户数）之前，代币表现多呈现中性或随大盘波动。
综合来看：短期更可能是情绪与行业新闻的“中性影响”，长期若预装钱包带来可观的真实用户增长、再叠加链上业务数据改善，才可能从中性转向偏多。