Space Bunny revealed: it was MiniMax M3.1 Flash

The mysterious "Space Bunny" stealth model that coding agents were buzzing about has a name. MiniMax has confirmed that Space Bunny was MiniMax M3.1 Flash — run anonymously to gather real-world usage before launch. Here is the full story, the specs, the price, and how to use the model today.

What happened

On 2026-09-23 an anonymous model called space-bunny-alphaappeared on OpenRouter as a "stealth" listing — a preview whose real provider is hidden while a lab gathers usage and feedback. It had a 1,000,000-token context window, multimodal input and standout coding, and quickly became a favorite to wire into OpenCode and other coding agents.

MiniMax has now revealed that Space Bunny was MiniMax M3.1 Flash. By MiniMax's own account the stealth run processed on the order of 200 trillion tokens of real traffic before the public launch — which is why the model felt so polished the moment it "arrived."

M3 or M3.1 Flash? Both, actually

Early on, reverse-engineering of the tokenizer and behavior pegged Space Bunny as MiniMax M3, and a named paid listing minimax/minimax-m3went live under that id. The reveal adds the missing detail: the specific stealth model people tested as "Space Bunny" was M3.1 Flash — the faster, cheaper sibling in the same M3 family. So both names are "right": M3 is the family and the full-size listing; M3.1 Flash is the exact stealth variant.

MiniMax M3.1 Flash — specs

• Family: MiniMax M3 (M3.1 Flash variant)
• Stealth name: Space Bunny (space-bunny-alpha on OpenRouter)
• Context window: 1,000,000 tokens
• Input: text, image, video — multimodal
• Type: reasoning model with tunable thinking depth
• Strengths: very fast inference, strong coding and agent tasks
• Launch price: $0.15 / 1M input, $0.60 / 1M output (cached input far cheaper)
• Open weights: expected 2026-10-23

How to use it today

You don't have to wait for the open weights or wrangle a provider account. Space Bunny runs on our hosted, OpenAI-compatible API — two models on one endpoint:

• space-bunny-flash — the M3.1 Flash tier, fast and cheap, $0.50 / 1M tokens
• space-bunny-alpha — the 1M-context flagship, $1.50 / 1M tokens (default)

Point any OpenAI SDK at https://spacebunnymodel.com/api/v1, send an sb_live_ key, and set the model field. Prepaid from $5, no subscription — or try it free in the browser playground first, no signup.