MiniMax Launches M3.1 Flash, Revealing Space Bunny Stealth Model

MiniMax on October 19, 2026, officially rolled out MiniMax M3.1 Flash, confirming in its launch post that Space Bunny, an anonymous model that processed more than 200 trillion tokens while operating in stealth, was an early version of the new release.

MiniMax’s announcement describes M3.1 Flash as a next-generation Flash model that delivers what the company calls a “generational leap” in agentic tasks, coding, computer use and cybersecurity. The company said the release makes online inference nearly three times faster while reducing usage costs by 70%.

The Space Bunny Stealth Run

MiniMax’s launch post said Space Bunny had topped global leaderboards while operating anonymously. OpenRouter, which hosted the stealth run, said in its reveal post that Space Bunny was its number-one model for 10 straight days and processed 56.7 trillion tokens, the most of any stealth model it has run.

OpenRouter said Space Bunny left its platform on October 5, 2026, and returned on October 19, 2026, as M3.1 Flash, which OpenRouter is offering free for 72 hours. OpenRouter credited feedback from users during the stealth period with helping MiniMax fix bugs and ship upgrades: time to first token improved 38%, from 1.95 seconds to 1.22 seconds, throughput rose 45%, from 70 tokens per second to 100 tokens per second, and the error rate fell from 1.4% to 0.7%.

Company-Reported Benchmark Results

According to MiniMax’s announcement, M3.1 Flash improves across a broad range of benchmarks, reaching what the company described as the level of leading models in complex agentic tasks, computer use, software development and cybersecurity.

MiniMax said the model ranks among the top performers on AutomationBench-AA and can complete real enterprise workflows across SaaS systems, solving practical problems in sales, marketing, operations, customer service, finance and HR from end to end. On BankerToolBench, which covers multi-step planning, maintaining consistency across documents and files, financial compliance and chained tool calls, the company reported a score of 76.2.

For coding and software engineering, MiniMax reported 73.7% on DeepSWE v1.1, a benchmark that focuses on complex software tasks requiring models to independently understand code, develop features, fix issues and validate their work through testing. The company said M3.1 Flash demonstrates strong execution across long sequences of steps, supporting complex software development workflows at a lower cost. The launch post cited DeepSWE at 74%, Automation-Bench at 62%, ALE at 30% and BabyVision, a visual-reasoning test, at 91%.

On computer use, MiniMax reported 61.7 points on OSWorld V2, which measures AI computer operation and cross-application task execution, and 29.6 points on Agents’ Last Exam (ALE), which evaluates performance on real professional work and complex tasks. The company said M3.1 Flash ranks first among Chinese models on both benchmarks, and described the model as capable of interacting with real environments, using tools and executing long sequences of tasks.

On cybersecurity, MiniMax reported 82% on CyberGym, which the company said demonstrates leading capabilities in vulnerability discovery and validation. The company added that in evaluations by its internal security experts, the model performed strongly in code auditing, vulnerability patching, penetration testing, malware analysis and reverse engineering.

OpenRouter’s post separately relayed MiniMax-reported results of 72.5% on AutomationBench-AA and 73.67% on DeepSWE, and said MiniMax’s own evaluations show M3.1 Flash edging DeepSeek V4.1 Flash on the Automation-Bench and HLE benchmarks.

MiniMax said it further optimized the model’s inference efficiency through more aggressive sparsity, reporting that output speeds can exceed 200 tokens per second in everyday use, with median (P50) speeds consistently above 170 tokens per second.

Pricing, Availability and the Weights Release

Flat API pricing is $0.15 per 1M input tokens, $0.60 per 1M output tokens and $0.015 per 1M cached input tokens. MiniMax said the model’s overall usage cost for agentic tasks is 70% lower than its predecessor, and the launch post described M3.1 Flash as delivering frontier-level intelligence across cowork, coding and visual-reasoning tasks at one-third the cost of M3 and three times the speed.

M3.1 Flash is available through the MiniMax API, the M Plan subscription and MiniMax Code for desktop. OpenRouter said that after its 72-hour free period, it will list the model at $0.15 per 1M input tokens, $0.60 per 1M output tokens and $0.015 per 1M cached input tokens.

MiniMax’s launch post said the model’s weights will arrive on October 23, 2026, while the company’s announcement said weights will be released later. MiniMax added that a Pro model, which it described as pushing the boundaries of intelligence, will also arrive later, and framed Flash as its ongoing exploration of the balance between efficiency and capability.