- A nameless 1-trillion-parameter AI model appeared on OpenRouter March 11 with zero attribution.
- It describes itself as "a Chinese AI model" with a May 2025 training cutoff identical to DeepSeek.
- Developers are split: is this DeepSeek V4 in disguise, or something else entirely?
On March 11, 2026, a ghost appeared in the machine. An AI model calling itself Hunter Alpha quietly showed up on OpenRouter, offering free access to a claimed 1-trillion-parameter system with a 1-million-token context window. No press release. No company name. No origin story.
The Specs That Stopped Everyone
Hunter Alpha's profile reads like a wishlist for AI developers:
- 1 trillion parameters — putting it in the same weight class as GPT-4-scale models
- 1,048,576 token context window — roughly 800,000 words, enough to process entire codebases or book manuscripts
- $0 per million input tokens, $0 per million output tokens — completely free
- Text-only input and output — optimized for agentic use cases
The combination caught immediate attention. "A trillion-parameter model with million-token context, available for free?" noted one developer on X. "That's not normal."
What Hunter Alpha Says About Itself
When Reuters put the chatbot through its paces, Hunter Alpha described itself as "a Chinese AI model primarily trained in Chinese" with a training data cutoff of May 2025. That cutoff date matches exactly what DeepSeek's own chatbot reports.
But ask who built it, and the response turns cryptic:
"I only know my name, my parameter scale and my context window length."
Neither DeepSeek nor OpenRouter has confirmed any connection. Both companies have ignored requests for comment.
Four Theories About Its Origin
The AI community has split into several camps:
Theory 1: DeepSeek V4 in Disguise
This is the leading theory. Hunter Alpha's specs align perfectly with leaked details about DeepSeek's upcoming V4 model. A "V4 Lite" variant briefly appeared on DeepSeek's own website days before Hunter Alpha launched, and DeepSeek has a history of anonymous releases followed by official announcements. The reported V4 launch timeline is April 2026. Counter-argument: some testers note Hunter Alpha's censorship is stronger and its math performance weaker than typical DeepSeek models.
Theory 2: Zhipu AI's GLM-6
The same anonymous OpenRouter provider account previously released "Pony Alpha," which later turned out to be GLM-5 from Chinese lab Zhipu AI. It's plausible Hunter Alpha is their next-gen text model (GLM-6) and its companion Healer Alpha is their new multimodal model.
Theory 3: OpenAI Test Build
OpenRouter has a documented history of hosting OpenAI test models under stealth names. Quasar Alpha became GPT-4.5 after testing. Hunter Alpha could follow that pattern.
Theory 4: A New Chinese Contender
Speculation points to Chinese labs Alibaba (Qwen), Tencent, or Baidu testing frontier systems through anonymous channels to gauge reception before official launch.
Performance: What It Does Well
Early testing reveals Hunter Alpha isn't a general-purpose chatbot. It's built for specific tasks:
- Agentic workflows: Tool-calling reliability is above average
- Instruction following: Rigid and accurate
- Long-context processing: Can handle massive documents
Weaknesses include creative tasks, complex mathematics, and general-purpose reasoning. Wharton professor Ethan Mollick tested it on the Lem Test for logical reasoning and called it "okay" — functional, not exceptional.
Why This Matters
Whether Hunter Alpha is DeepSeek V4, GLM-6, or something else, its appearance signals a shift. Chinese AI labs are releasing frontier-scale models through anonymous channels, offering them for free, and letting the community benchmark them before any official announcement.
The model is live. The identity remains unknown. And a community hub — hunteralphahub.com — has already launched to track clues.
Read more about China's AI agent investments and Nvidia's GTC 2026 announcements that are reshaping the AI landscape.
FAQ
Is Hunter Alpha free to use?
Yes, Hunter Alpha is currently 100% free on OpenRouter with no costs for prompts or completions. This has raised questions about whether it's a limited-time testing phase before monetization.
How do I access Hunter Alpha?
Create an account at OpenRouter.ai, search for Hunter Alpha, and use it via API or compatible interfaces like OpenClaw and other agentic frameworks.
What makes Hunter Alpha different from other models?
The combination of 1T parameters, 1M context window, and free access is unprecedented. Most models with these specs cost tens of dollars per million tokens.
CTA: Want to build AI agents that use models like Hunter Alpha? Check out how developers are building agentic pipelines and the latest in LLM tooling.