MiniMax M3.1 Flash was introduced and the secret Space Bunny model has been revealed

MiniMax officially released the M3.1 Flash model on October 19, 2026, and confirmed that Space Bunny, which had been competing anonymously for weeks, was an early version of this model. According to the company, Space Bunny reached a processing volume of more than 200 trillion tokens during its hidden period.

It stayed on the leaderboard for ten days

OpenRouter, which hosted Space Bunny, said the model was the platform’s number one for 10 consecutive days and processed 56.7 trillion tokens on its own infrastructure. This figure is the highest among the hidden models the platform has run to date. The model left the platform on October 5 and returned on October 19 as M3.1 Flash. OpenRouter is offering the model for free for 72 hours.

User feedback went directly into the model

According to data shared by OpenRouter, feedback received during the hidden period helped MiniMax fix bugs and upgrade the release. Time to first token dropped from 1.95 seconds to 1.22 seconds, a 38% improvement. Generation speed rose from 70 tokens per second to 100 tokens per second, and the error rate fell from 1.4% to 0.7%.

MiniMax says online inference for M3.1 Flash is about three times faster and usage costs have dropped by 70%. According to the company, the model can exceed 200 tokens per second in daily use, with median speed staying above 170 tokens. MiniMax describes the new release as a generational leap in agent tasks, coding, computer use, and cybersecurity.

Benchmark results and pricing

According to the results MiniMax announced, the model scored 73.7% on the DeepSWE v1.1 test, 76.2 points on BankerToolBench, 61.7 points on OSWorld V2, and 29.6 points on Agents’ Last Exam. It recorded 82% on the cybersecurity test CyberGym and 91% on the visual reasoning test BabyVision. These figures are the company’s own evaluations and need to be verified by independent tests.

API pricing is $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 for cached input. The model is accessible through the MiniMax API, the M Plan subscription, and MiniMax Code for desktop. The weights are expected to be released on October 23, 2026, and a more powerful Pro version will arrive later.

Is the hidden model era ending?

The Space Bunny case shows that anonymously released models are no longer just curiosity-driven experiments but planned launch tools. Users test the model without knowing it, feedback is collected, and the model returns with an identity and pricing. For MiniMax, this strategy worked. The open-weight promise will be tested on October 23.

Source: Unite.AI