Free stealth model Space Bunny Alpha draws coders, maker unnamed

A model called Space Bunny Alpha appeared on OpenRouter on Tuesday at no charge, and builders began posting what they had made with it through Wednesday and Thursday. OpenRouter's listing gives it a one-million-token context window, a maximum of 524,288 completion tokens, and text, image and video input. It carries no price for prompt or completion tokens.

OpenRouter describes it as "an anonymous large model with blazing-fast inference, strong coding capabilities and native multimodal input support", and says the provider operating it has chosen to remain anonymous during the preview. That description is the provider's own. Nobody outside it has published a benchmark run, and the listing shows the model is not moderated.

The account mark_k said on Thursday that community fingerprinting points to a MiniMax M3.1 preview, and added that nobody has officially confirmed who made it. The post had 71 likes when read at 18:00 UTC on Thursday. MiniMax has not said anything publicly about the model, and OpenRouter's page names no lab.

The demos are what is carrying it. Atul Kumar, who has 141,000 followers, said OpenCode had made Space Bunny free for a week and that he used it to build a portfolio site with code-generated animations; the post drew 98 likes and 43,600 views. Vox, an account that posts model tests, said it produced a 1,564-line GPU fluid simulation in a single HTML file with no libraries.

Both are single accounts reporting their own results, and neither has published the prompts or the code for anyone else to check. The account polydao put first-token latency at about 1.14 seconds in its own runs. OpenRouter's listing says the model requires reasoning, with effort settings from low through max.

Stealth listings of this kind usually end with a lab claiming the model or the endpoint going dark. Neither has happened yet.