Space Bunny AI Was MiniMax's Stealth Model All Along

Space Bunny AI was never a mystery competitor at all. On October 19, 2026, the Chinese AI lab MiniMax confirmed that the pseudonymous model quietly topping public leaderboards was an early build of its own newly launched flagship — and that it had processed more than 200 trillion tokens under the alias before going public, according to Tech Insider.

The admission ends months of guessing among AI developers about who built the high-scoring mystery entry and why it vanished without ever being named. MiniMax folded the answer directly into the launch of its new model, turning anonymous leaderboard dominance into a ready-made marketing story on day one. It lands in a week when rival labs are racing to ship cheaper, faster agentic models of their own.

What the model actually was

According to the company's own account, Space Bunny was simply an earlier version of the production model now shipping under the official identifier MiniMax-M3.1-Flash. That framing matters: developers who had been testing and benchmarking the mystery model for weeks were not evaluating an experimental side project or a leaked competitor build. They were using the flagship in disguise.

Stealth testing of this kind is not unusual in the current AI model race. Labs frequently push unlabeled or pseudonymously named checkpoints onto public inference marketplaces to gather real-world usage data and leaderboard rankings before a formal launch, reducing the risk of a public benchmark miss. What is less common is a company explicitly confirming the identity of a stealth model after the fact, rather than letting the connection remain speculation. MiniMax did exactly that, tying the alias's token count and leaderboard standing directly to the production model it just released through its API.

The new system is positioned as an agentic model built for tool use, coding, computer-use tasks, cybersecurity workflows, and multimodal input. It offers a context window stretching up to one million tokens and a thinking-depth setting that developers can tune for a given task — lighter reasoning for high-volume, low-latency agent calls, and deeper reasoning when a task genuinely needs it. For teams running thousands of automated agent calls a day, that flexibility can matter more than any single flashy benchmark number.

Benchmarks and price

MiniMax published four headline benchmark figures alongside the launch. On Automation-Bench, which measures multi-step automated workflows, the model scored 62 percent. On ALE, a tougher agentic-reasoning exam, it scored 30 percent. On DeepSWE, a software-engineering benchmark, it reached 74 percent. And on BabyVision, a visual-reasoning test, it hit 91 percent.

The gap between the first two scores is worth noting on its own. A strong result on a general automation benchmark next to a modest one on a test built to stress-test agentic reasoning suggests the model is more reliable at executing defined multi-step workflows than at open-ended, exam-style problem solving — a useful distinction for any team deciding whether it fits a structured automation pipeline.

Pricing is set at about fifteen cents per million input tokens and sixty cents per million output tokens through the MiniMax API, with cached input billed at roughly a tenth of that. Combined with the million-token context window, the structure is designed for workloads that repeatedly reuse large context blocks, such as an agent that keeps re-reading the same codebase or document set across many calls. The company's headline performance claim is that the successor to the Space Bunny AI prototype runs nearly three times faster in online inference than its predecessor while cutting usage costs by roughly seventy percent.

MiniMax initially said model weights would follow the hosted launch at a later date, then set that date as October 23, 2026 — four days after the model went live. That gap between a hosted launch and an open-weight release has become a familiar pattern: ship the API first to capture usage and benchmark attention, then follow with weights once the rollout has stabilized. For the open-source community, the weights release matters because it lets independent researchers verify the company's benchmark claims directly and build fine-tuned variants of their own.

Why stealth reveals matter

The launch arrives in a crowded field of low-cost, high-context Chinese models built for agentic and coding work. DeepSeek updated its own flash-tier model under an MIT license just days before MiniMax's announcement and posted a DeepSWE score of 77, putting pressure on MiniMax's 74. Reports have framed the two launches as part of the same competitive wave, with MiniMax, DeepSeek, GLM, and Kimi all shipping fast, agent-focused models within weeks of each other — a cadence that has made speed and price the differentiators developers can act on immediately.

For enterprise buyers, the Space Bunny AI reveal carries a warning label. A model sitting near the top of a public leaderboard under an unfamiliar name may well be a major lab's unreleased flagship rather than an independent newcomer, which changes how much weight that ranking should carry in a procurement decision. Teams that track leaderboard standings as part of vendor evaluation now have fresh reason to ask who actually built a high-ranking anonymous model before assuming it reflects a smaller, less-resourced competitor.

The episode also shows how leaderboard strategy has become launch marketing. MiniMax turned months of anonymous activity into evidence of real-world readiness, a concrete talking point that a benchmark chart alone cannot provide — and one that arrives as AI agents reshape how brands get discovered, AI startups win new backing, and the industry's competitive stakes keep rising, as seen in recent AI industry courtroom drama.