Economy March 18, 2026

Xiaomi Identified as Source of Anonymous 'Hunter Alpha' AI Model Previously Attributed to DeepSeek

MiMo team confirms Hunter Alpha was an internal test build of MiMo-V2-Pro as debate around next-generation models and market implications continues

By Ajmal Hussain
Xiaomi Identified as Source of Anonymous 'Hunter Alpha' AI Model Previously Attributed to DeepSeek

A powerful unnamed AI model that appeared anonymously on OpenRouter has been confirmed by Xiaomi's MiMo team to be an early internal test build of MiMo-V2-Pro, ending speculation that it was the long-awaited DeepSeek-V4. The model, labeled Hunter Alpha on the platform, drew attention for its advertised one-trillion-parameter size and a one-million-token context window, and was rapidly adopted on OpenRouter before Xiaomi acknowledged ownership.

Key Points

  • Xiaomi's MiMo team confirmed Hunter Alpha was an internal test build of MiMo-V2-Pro, ending speculation that the anonymous model was DeepSeek-V4.
  • Hunter Alpha advertised a 1-trillion-parameter scale and a one-million-token context window, features that attracted rapid adoption and heavy testing on OpenRouter.
  • The incident underscores interactions between stealth model testing on gateway platforms, rapid user adoption, and ongoing interest in agent-focused AI tooling such as OpenClaw; market responses previously included a global tech stock selloff tied to DeepSeek's low-cost models.

A previously uncredited artificial intelligence model that surfaced on March 11 on the OpenRouter developer gateway has been confirmed as coming from Chinese technology firm Xiaomi, the company's MiMo AI team said on Wednesday. The model, which had appeared anonymously under the name Hunter Alpha, prompted industry and market speculation that it might be the much-anticipated DeepSeek-V4.

MiMo, the AI model group at Xiaomi led by former DeepSeek researcher Luo Fuli, described Hunter Alpha as "an early internal test build of MiMo-V2-Pro." The company said the model is intended to function as the "brain" for AI agents - software tools that aim to let users accomplish complex tasks with fewer prompts and less human oversight compared with traditional chatbots.


Context and market reaction

Interest in an advanced DeepSeek model has been high since the low-cost releases of DeepSeek-V3 and R1 last year, a development that triggered a global selloff in technology shares and led investors to question whether U.S. AI companies needed to continue spending heavily on large-scale AI computing resources. Expectations that DeepSeek-V4 could be a next-generation leap have featured prominently in local reporting, with some outlets suggesting a launch could follow soon.

Those expectations helped fuel speculation when Hunter Alpha appeared without developer attribution on OpenRouter. The anonymous listing, later labeled a "stealth model" by the platform, matched some public assumptions about the capabilities such a next-generation model might display.


Technical profile and on-platform behavior

During Reuters tests, the chatbot operating as Hunter Alpha described itself as "a Chinese AI model primarily trained in Chinese" and indicated its training data extended to May 2025 - the same knowledge cutoff date reported for DeepSeek's own chatbot. When pressed about its origin, the system declined to name its developer, stating: "I only know my name, my parameter scale and my context window length."

Hunter Alpha's publicly visible profile on OpenRouter listed it as a 1-trillion-parameter model with a context window of up to one million tokens. A token is an element of text - roughly part of a word - that an AI model consumes when processing input and generating output. The combination of a very large context window and advertised reasoning capability, together with free access on the platform, attracted attention from engineers and developers testing agent systems.

"The combination that stood out was Hunter Alpha's 1-million-token context paired with reasoning capability and free access," said Nabil Haouam, an engineer who builds AI agent systems. Haouam noted that most frontier models offering that scale of context typically incur significant costs when used at scale.


Stealth launches and testing practices

Anonymous or uncredited model appearances are not uncommon on gateway services like OpenRouter, which let developers route queries to many AI systems via a single interface. Such platforms are often used to trial models without formally announcing them. A prior example on OpenRouter involved a model named Pony Alpha in February that was later confirmed as part of Zhipu AI's GLM-5 system.

Hunter Alpha's OpenRouter profile also included a notice that all prompts and completions "are logged by the provider and may be used to improve the model," a disclosure that underscores an industry practice of using stealth launches to collect unbiased feedback from users and developers.


Adoption on OpenRouter

Following its appearance, Hunter Alpha saw rapid uptake on the platform. According to Xiaomi's MiMo team, the model surpassed one trillion tokens in total usage and reached top positions on OpenRouter's leaderboard prior to Xiaomi's acknowledgement that the build was MiMo-V2-Pro in early internal testing.

Observers who run independent benchmark tests said the timing and advertised capabilities of Hunter Alpha made conjecture linking it to DeepSeek understandable. Umur Ozkul, an independent tester, noted that the convergence of timing and the model's stated specifications helped explain why users and commentators associated the anonymous model with next-generation offerings from DeepSeek.


Broader technical ecosystem

Xiaomi's confirmation comes amid rapidly expanding interest in open-source agent frameworks in China, including growing adoption of OpenClaw, which enables a wide range of users to deploy agent-style tools. The development highlights how stealth testing, gateway platforms, and agent frameworks are intersecting as companies iterate on models intended to power more autonomous, task-oriented AI experiences.

Risks

  • Ambiguous developer attribution for anonymous models can generate market speculation and uncertainty affecting technology investor sentiment - particularly in the AI and broader tech sectors.
  • Stealth testing practices that log prompts and completions - as disclosed on Hunter Alpha's profile - create uncertainty for users about how their inputs may be used to improve models, which could influence adoption among enterprise and consumer users.
  • Rapid, unsanctioned adoption of uncredited models on public gateways may complicate benchmarking and attribution, posing challenges for competitors, researchers, and downstream platforms in assessing capabilities and provenance.

More from Economy

China Likely to Hold Lending Rates as Middle East Conflict Clouds Inflation Outlook Mar 19, 2026 Trump Says U.S. Could 'Blow Up' South Pars If Iran Strikes Qatar's LNG Again Mar 19, 2026 Lower House Confirms Takaichi's Dovish Picks to BOJ Board, Potentially Shaping Rate Path Mar 18, 2026 BOJ Keeps Rates on Hold, Flags Middle East Oil Risks to Inflation Mar 18, 2026 BOJ Holds Rates at 0.75% and Flags Oil-Driven Inflation Risks Mar 18, 2026