OpenRouter platform recently added two mysterious new models: Hunter Alpha and Healer Alpha, sparking community discussions.

According to leaks, Hunter Alpha has an impressive 1 trillion (1T) parameters and a 100,000 token long context, supports multimodal input (but outputs text), and has an output speed of about 48 tokens/s. This model is optimized for agent (agentic) scenarios, excelling in long-term planning, complex reasoning, and continuous multi-step task execution, and is positioned as a cutting-edge tool for agent workflows.

QQ20260312-111616.jpg

The other model, Healer Alpha, has a context window of 262K tokens (about 262,000), with undisclosed parameters. However, based on its output speed of up to 93 tokens/s, it is speculated that its activated parameter count may be about half that of Hunter Alpha. This model also supports multimodal input, emphasizing visual and audio understanding and real-world execution capabilities, and is described as a cutting-edge omni-modal model with vision, hearing, reasoning, and action capabilities.

Currently, both models are launched in stealth mode, from the same undisclosed provider, and are temporarily free to use (all prompts and completions will be recorded for model improvement).

QQ20260312-111623.jpg

Some netizens speculate whether it is DeepSeek-V4, but after analyzing the provider's historical records, it is more likely to be the new generation work from ZhiPu AI (Z.ai). Previously, the same provider anonymously released Pony Alpha, which was later confirmed to be GLM-5. Therefore, the community generally speculates:

  • Hunter Alpha is likely to correspond to ZhiPu's new flagship text model (parameters jump to the 1T level)
  • Healer Alpha may be GLM-5V or a new omni-model multimodal version

At present, the official has not confirmed the identity, and the specific performance and ownership remain to be further verified. Let's wait and see what will be revealed next!