Anonymous Model 'Space Bunny' Tops OpenRouter API Call Rankings; Community Evidence Points to MiniMax's New Product
nashnova research
Anonymous large language model Space Bunny hit No. 1 in daily token volume on both OpenRouter and OpenCode within three days of launch — roughly 3.99 trillion and 8.3 trillion tokens respectively — while independent tokenizer tests point to MiniMax's M3.1 Flash as the likely creator.
How does a nameless model top two platforms in three days?
Space Bunny launched September 23. By September 27 it had held the No. 1 daily usage spot for three straight days on both OpenRouter and OpenCode.
Token volume: roughly 3.99 trillion on OpenRouter, 8.3 trillion on OpenCode — the two main gateways where developers worldwide test and call LLMs.
This means → developers adopted it at massive scale without knowing who built it. The product itself did the convincing.
How did the community figure out it might be MiniMax?
Tokenizer testing was the key evidence. A tokenizer is the tool that chops text into the smallest units a model can read. A Reddit user sent seven different strings to Space Bunny, logged the token counts, and compared them against public tokenizers from multiple model families. MiniMax's M-series matched perfectly across all seven.
Independent research site YFarmX ran 50 test strings — every one matched MiniMax's tokenizer output.
Developer Lumina found traces of M3.1 inside MiniMax Code's file and model directories; the 1-million-token context window and Max reasoning-tier config also matched Space Bunny's specs.
In plain terms = three independent testers used different methods and reached the same conclusion — like lifting fingerprints at different scenes and getting the same match every time.
What do developers say after actually using it?
Developer Siddharth fed Space Bunny a prompt and a few images to generate a seaside browser game called Seaside. The demo video drew over 80,000 views across platforms.
Developer Alan tested four coding tasks — a FastAPI endpoint, Terraform-based AWS infra, a Stripe-style data dashboard, and a ChromaDB data pipeline. He reported all four passed on the first generation with no edits or debugging needed.
In Reddit comparisons, users rated Space Bunny's web-design output as slightly better than GLM-5.3 Flash, with stronger instruction-following, fewer hallucinations, and faster speed than MiMo 2.6. Another user called its contextual understanding the best among peers, with reasoning steps that mostly added real value.
What does the Flash-model landscape look like now?
The latest OpenRouter leaderboard top three: Space Bunny, DeepSeek V4.1 Flash, and Zhipu GLM 5.3 Flash — forming a new Flash-tier "big three."
Every major vendor is converging on the same playbook: stronger coding, agent, and multimodal capabilities while pushing down latency and per-call cost.
Google updated its Flash line three times in six weeks. DeepSeek, Zhipu, and Xiaomi's MiMo have all shipped Flash variants in quick succession.
This means → the Flash tier has shifted from "cheap fallback" to the main battlefield. The cost-performance war for high-frequency calling scenarios is now fully underway.
How does the identity mystery end?
Space Bunny's real development team remains officially unconfirmed.
If MiniMax (HKEX: 00100) eventually claims it, its competitive standing in high-frequency model markets gets direct validation — anonymous real-world usage data proving product strength.
If it turns out to be another team, this signals that the Flash tier has more contenders than the market expected, and the competitive map is far from settled.
市场有风险，内容仅供研究参考，不构成投资建议。