Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On Radar3DNews AI→ original

Xiaomi reveals MiMo-V2-Pro: Hunter Alpha was not a new DeepSeek model

Xiaomi has revealed the origin of the mysterious Hunter Alpha on OpenRouter: it was an early version of MiMo-V2-Pro, the company's flagship model for AI agents. The model has more than 1 trillion parameters, a 1-million-token context window and support for OpenClaw. Before the announcement, many assumed it was a hidden test of DeepSeek V4, so strong did the model look against the rest of the market. Xiaomi says MiMo-V2-Pro is built for real-world agentic workloads rather than chat alone.

AI-processed from 3DNews AI; edited by Hamidun News
Xiaomi reveals MiMo-V2-Pro: Hunter Alpha was not a new DeepSeek model
Source: 3DNews AI. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary source3DNews AI ↗
Editor-in-Chief
Key metric #1More than 1 trillion total parameters
Key metric #242 billion active parameters
Key metric #31-million-token context window

Xiaomi has officially revealed the origin of the mysterious Hunter Alpha model, which appeared on OpenRouter on March 11, 2026 with no developer named. It turns out MiMo-V2-Pro, the company's flagship model for agentic scenarios, was being tested under that name, and its specs are why the market suspected a new DeepSeek.

How the model was unmasked

When Hunter Alpha was first spotted on OpenRouter, it immediately set off a wave of discussion. The model described itself as Chinese, said it had been trained mainly on Chinese-language data, and gave a knowledge cutoff of May 2025. Asked who built it, it declined to answer, saying only that it knew its own name, its parameter scale and its context window size. Amid the anticipation around DeepSeek V4, that was enough for rumors to spread across the market within days.

On March 18, 2026, the Xiaomi MiMo team said Hunter Alpha was an early internal test build of MiMo-V2-Pro. The team's makeup added to the intrigue: the MiMo effort is led by Luo Fuli, a former DeepSeek engineer. In her statement she described the moment as a sharp shift from ordinary chats to agentic systems, meaning models that do not just answer but can carry out chains of actions through tools and external services.

"The shift from chats to agents happened so fast that even we could barely believe it."

What MiMo-V2-Pro can do

Xiaomi positions MiMo-V2-Pro as a base model for real agentic workloads. The idea is that such a system should be more than a conversation partner: it should act as the "brain" of agent infrastructure, coordinating complex workflows, holding long context, calling tools and reliably seeing a task through to a result. The emphasis is therefore not only on general language ability but on practical behavior in scenarios where a model must plan, choose steps and act sequentially.

  • More than 1 trillion total parameters
  • 42 billion active parameters
  • 1-million-token context window
  • Support for five agent frameworks, including OpenClaw
  • Free trial access for developers for one week after the announcement

In the Artificial Analysis ranking, the model placed eighth globally and second among Chinese language models. Xiaomi also claims that MiMo-V2-Pro beats Claude 4.6 Sonnet at code generation and approaches Opus 4.6 in overall performance on agentic tasks. The company also stresses that the new version is markedly more stable and more accurate at calling tools than the previous lineup.

Why everyone thought of DeepSeek

The theory that Hunter Alpha might be a new DeepSeek did not come out of nowhere. After the release of DeepSeek-V3 and DeepSeek-R1, the market began watching Chinese labs much more closely: cheap, strong models noticeably shook investor expectations and raised the question of whether the industry really needs its previous scale of compute spending. Against that backdrop, an anonymous reasoning model with a million tokens of context and strong performance on agentic tasks looked like a plausible hidden test of DeepSeek V4.

The launch format also played a part. Platforms like OpenRouter have long been a convenient venue for quiet trials: developers get a live stream of requests, and users test the model without preconceptions about the brand. According to MiMo, Hunter Alpha quickly passed 1 trillion tokens of usage and reached the top of the OpenRouter rankings. After Xiaomi officially disclosed the model's origin, its shares in Hong Kong briefly gained as much as 5.8%.

What it means

The MiMo-V2-Pro story shows that the new competition is no longer only about chatbots but about agentic models that can use tools and run long chains of tasks. For Xiaomi, it is a bid to be seen not just as a maker of smartphones and electric cars but as a provider of its own AI platform. For the market, it is another signal: strong models can now come not only from startups like DeepSeek but also from large consumer ecosystems.

Story

Key Questions Answered

How the model was unmasked?

When Hunter Alpha was first spotted on OpenRouter, it immediately set off a wave of discussion. The model described itself as Chinese, said it had been trained mainly on Chinese-language data, and gave a knowledge cutoff of May 2025. Asked who built it, it declined to answer, saying only that it knew its own name, its parameter scale and its context window size. Amid the anticipation around DeepSeek V4, that was enough for rumors to spread across the market within days.

What MiMo-V2-Pro can do?

Xiaomi positions MiMo-V2-Pro as a base model for real agentic workloads. The idea is that such a system should be more than a conversation partner: it should act as the "brain" of agent infrastructure, coordinating complex workflows, holding long context, calling tools and reliably seeing a task through to a result. The emphasis is therefore not only on general language ability but on practical behavior in scenarios where a model must plan, choose steps and act sequentially.

Why everyone thought of DeepSeek?

The theory that Hunter Alpha might be a new DeepSeek did not come out of nowhere. After the release of DeepSeek-V3 and DeepSeek-R1, the market began watching Chinese labs much more closely: cheap, strong models noticeably shook investor expectations and raised the question of whether the industry really needs its previous scale of compute spending.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…