MiniMax M3.1 Flash

Today, we officially rotate out MiniMax M3.1 Flash.

Space Bunny 🐰, which processed additional than 200 trillion tokens during functioning anonymously, was an first type of MiniMax M3.1 Flash.

As a next-generation Flash model, M3.1 Flash delivers a generational jump in agentic tasks, coding, Computer Use and cybersecurity, during making online conclusion nearly three times faster and reducing use expenses by 70%.

M3.1 Flash improves throughout a broad range of benchmarks, reaching the flat of foremost models in complex agentic tasks (cowork), device use, application betterment (coding), and cybersecurity.

Complex Tasks and Automation:

• M3.1 Flash ranks among the top performers on AutomationBench-AA. It can complete genuine endeavor workflows throughout SaaS systems, solving applicable problems in sales, marketing, operations, client service, finance, and HR from end to end.
• Scoring 76.2 on BankerToolBench, it excels at complex real-world challenges specified as multi-step planning, maintaining consistency throughout documents and files, financial compliance, and chained tool calls.

Coding and Software Engineering:

• It achieves 73.7% on DeepSWE v1.1, placing it alongside top models. DeepSWE focuses on complex application tasks that necessitate models to independently comprehend code, create features, fix issues, and validate their activity through testing. M3.1 Flash demonstrates powerful implementation throughout lengthy sequences of steps, supporting complex application betterment workflows at a lesser cost.

Computer Use:

• It scores 61.7 points on OSWorld V2, which measures AI device procedure and cross-application project execution, and 29.6 points on Agents’ Last Exam (ALE), which evaluates achievement on genuine expert activity and complex tasks. It ranks archetypal among Chinese models on the two benchmarks.
• M3.1 Flash demonstrates powerful capabilities in interacting alongside genuine environments, using tools, and executing lengthy sequences of tasks—moving beyond understanding a project to taking act and delivering results.

Cybersecurity:

• It achieves a powerful mark of 82% on CyberGym, demonstrating foremost capabilities in exposure finding and validation.
• In in-depth evaluations by our inner safety experts, the example performed powerfully in code auditing, exposure patching, intrusion testing, malware analysis, and reverse engineering, enabling efficient, low-cost assistance for construction complex safety systems.

We have additional optimized M3.1 Flash’s conclusion effectiveness through additional assured sparsity. In mundane use, output speeds can attain 200+ TPS (tokens per second), alongside median (P50) speeds consistently reaching 170+ TPS.

For agentic tasks, the model’s general use disbursal is 70% lesser than its predecessor.

Flat pricing per 1M tokens: $0.15 input · $0.60 output · $0.015 cached input.

Across three benchmarks—Automation-Bench for complex agentic project execution, DeepSWE for real-world coding, and Agents’ Last Exam (ALE) for device use in real-world tasks alongside financial value—M3.1 Flash delivers achievement on par alongside top models at a extremely low project cost, environment a extremely rivalrous new norm for achievement and cost.

M3.1 Flash is now accessible through the MiniMax API, M Plan, and MiniMax Code. Model weights volition be released later.

Download MiniMax Code for desktop: https://agent.minimax.io/download

Subscribe to M Plan: https://www.minimax.io/m-plan

API access: platform.minimax.io

Flash represents our ongoing exploration of the balance between effectiveness and capability. Pro, which pushes the boundaries of intelligence, volition additionally attain later.

Intelligence alongside Everyone.