I used the subtitle translation skill to test subtitle translation on Space Bunny. The skill calls multiple tools and incorporates various rules. I generally use it to conduct “drooling” tests on new models.
Those interested can test it themselves
Verdict: Fast and effective—definitely in the non-drooling category.
I just saw this in workbuddy: 0.03x. Not sure what it is.
“Space Bunny” currently primarily refers to an anonymous large language model launched on September 23, 2026, known in Chinese-language communities as “Space Rabbit.” It was released by an unidentified third-party provider on the OpenRouter and OpenCode platforms, focusing on fast reasoning and coding capabilities.
Core Specifications
· Context window: 1 million tokens, with native multimodal input support (text, images, and video).
· Output and capabilities: Maximum output of approximately 524,000 tokens, with tool calling and adjustable reasoning intensity.
· Current status: Free for a limited time, but the provider’s identity, parameter count, and architecture have not been disclosed.
Why It Has Attracted Attention
Just three days after launch, it topped the daily usage rankings on OpenRouter and OpenCode, remaining at the top consecutively. Developers who tested it found that its coding and visual-generation performance was close to Claude Opus 5, while it was faster and could even proactively take screenshots to inspect and fix issues in pages it had generated.
Guesses About Its Identity
The community discovered through Tokenizer fingerprint tests that its token counts matched those of MiniMax models exactly across 50 test strings. It is widely speculated to be an anonymous preview version of MiniMax (such as M3.1), but the company has not officially confirmed this.
I’ve already started using it. Is the reasoning still a bit weak?
