Overview
MiniMax M3.1 Flash is a multimodal MoE model released by MiniMax on October 19, 2026, namely the model previously anonymously tested on OpenRouter under the codename Space Bunny. It has 428B total parameters and about 23B activated parameters, adopts sparse attention and a native vision encoder, supports text, image, and video inputs, outputs text, and has a context window of 1,000,000 tokens. The model focuses on coding and agent workflows, priced at $0.15 per million input tokens and $0.60 per million output tokens, with no extra charge for long prompts, roughly half of the M3 standard rate, and the company says it is about three times faster than M3. During anonymous testing, Space Bunny ranked first on OpenRouter for 10 consecutive days, processing 56.7T tokens, and ranked first on OpenCode for 12 consecutive days, accumulating 106T tokens, processing over 200T tokens in total.
Key Features
- Multimodal MoE architecture: A sparse MoE model with 428B total parameters and about 23B activated, equipped with a native vision encoder, supporting text, image, and video inputs and outputting text.
- Million-level context window: The context window reaches 1,000,000 tokens, with no extra charge for long prompts, suitable for handling large-scale codebases and long-document tasks.
- Always-on reasoning and five effort levels: Reasoning is always on and cannot be disabled, offering five effort levels: Low, Medium, High, Xhigh, and Max, with Max officially recommended; reasoning content is returned through a separate reasoning_content field.
- Dual API format compatibility: Compatible with both Anthropic and OpenAI API formats, usable through the MiniMax API, M Plan subscription, and the MiniMax Code desktop app.
- Positioned for coding and agent workflows: Focused on coding and agent workflows, with official benchmarks covering coding and agent evaluation items such as Automation-Bench v1.0.6, DeepSWE v1.1, OSWorld V2, and CyberGym.
- Broad platform availability: Launched on OpenRouter (72 hours free), OpenCode, Ollama, Cline, and GMI Cloud on release day, with model weights planned for release on October 23.
Use Cases
- Large-scale codebase understanding and coding assistance, leveraging the 1M context window to handle long code files and cross-file dependencies.
- Building agent workflows, using always-on reasoning and multiple effort levels to adapt to automated tasks of varying complexity.
- Multimodal content analysis, taking image or video inputs and combining them with text for understanding and summarization.
- Long-document and long-prompt processing, such as contract review and research report analysis, which require large contexts.
- Quick integration and trial through platforms such as OpenRouter, OpenCode, Ollama, Cline, and GMI Cloud.
Pros
- Priced at $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 for cache reads, with no extra charge for long prompts, roughly half of the M3 standard rate.
- The company says it is about three times faster than M3, suitable for coding and agent scenarios with throughput requirements.
- During anonymous testing, it ranked first on OpenRouter for 10 consecutive days, processing 56.7T tokens, and ranked first on OpenCode for 12 consecutive days, accumulating 106T tokens, with Space Bunny processing over 200T tokens in total.
- Provides a 1,000,000-token context window and supports multimodal inputs of text, image, and video.
- Compatible with both Anthropic and OpenAI API formats, launched on multiple mainstream platforms on release day, with flexible integration options.
- Reasoning effort offers five adjustable levels, with Max officially recommended, making it easy to balance reasoning depth according to task complexity.
Pricing
$0.15 per million input tokens, $0.60 per million output tokens, and $0.015 for cache reads, with no extra charge for long prompts, roughly half of the M3 standard rate. Usable through the MiniMax API, M Plan subscription, and the MiniMax Code desktop app, with 72 hours free on OpenRouter on release day. Other billing details are subject to the official website.
Summary
MiniMax M3.1 Flash is a multimodal MoE model focused on coding and agent workflows, featuring 428B total parameters, about 23B activated, a 1M context, and five reasoning effort levels, priced at roughly half of the M3 standard rate. During anonymous testing it achieved consecutive first-place rankings on OpenRouter and OpenCode, launched on multiple platforms on release day, and is suitable for coding and agent tasks requiring long context and multimodal input.
Version History
- MiniMax M3.1 Flash released: 428B MoE with 1M context at $0.15 per million input tokens (2026-10-19): MiniMax released M3.1 Flash on 2026-10-19, revealing it as the model previously tested anonymously on OpenRouter under the codename Space Bunny. It is a multimodal MoE with 428B total parameters and about 23B active, a 1M context window, and support for text, image and video input with text output. Pricing is $0.15 per million input tokens, $0.60 for output and $0.015 for cached reads, with no surcharge for long prompts. Reasoning is always on and cannot be disabled, with five effort levels: Low, Medium, High, Xhigh and Max. It shipped the same day on OpenRouter (free for 72 hours), OpenCode, Ollama, Cline and GMI Cloud, with weights scheduled for October 23.