MiniMax M3.1 Flash
MiniMax M3.1 Flash is a next-generation flagship multimodal MoE model engineered for high-performance agentic automation, complex code synthesis, and deep reasoning. Built on a massive 428B parameter Mixture-of-Experts architecture and accelerated by MiniMax Sparse Attention (MSA), it accelerates inference speeds while expanding context throughput up to 1 million tokens.
Related Models
MiniMax M2.7
MiniMax-M2.7 is a next-generation large language model built for autonomous real-world productivity and continuous improvement. Through advanced agentic capabilities and multi-agent collaboration, it can plan, execute, evaluate, and refine complex tasks in dynamic environments while actively contributing to its own evolution. Optimized for production-grade workflows, M2.7 excels at live debugging, root-cause analysis, financial modeling, and end-to-end document creation across Word, Excel, and PowerPoint. It achieves 56.2% on SWE-Pro, 57.0% on Terminal Bench 2, and a 1495 ELO rating on GDPval-AA, establishing a new benchmark for multi-agent systems operating in real-world digital workflows.
MiniMax M3
MiniMax-M3 is MiniMax’s latest M-series multimodal foundation model, built for agentic reasoning, tool use, coding, and long-context tasks. It supports text, image, and video inputs with text output, offering a 1M-token context window, extended thinking, function calling, and structured outputs. With strong capabilities in long-horizon agent workflows, software development, multimodal understanding, and extended response generation, MiniMax-M3 is ideal for autonomous agents, coding assistants, document and video analysis, and production-grade applications that require massive context at a competitive cost.
Read Me
MiniMax M3.1 Flash API
MiniMax M3.1 Flash is a high-throughput, low-latency multimodal large language model developed by MiniMax, corresponding to the official M3.1-Flash-Preview released on September 27, 2026. It is designed for ultra-fast Agent responses, code generation and analysis, massive long-context throughput, and high-concurrency multimodal tasks. It accepts text, image, and video inputs while outputting text.
On the iCreat platform, MiniMax M3.1 Flash is provided as an independent model code with pay-as-you-go pricing. The endpoint is OpenAI-compatible Chat Completions at https://api.icreat.ai/v1, supporting a 1-million-token context window and a 524,288-token output limit. Pricing is $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cache-read tokens, approximately half of the official M3 pay-as-you-go rates.
Model Positioning
MiniMax M3.1 Flash is positioned as a fast, reliable model for everyday development, ranging from quick bug fixes to full-feature implementation. Compared to the previous-generation MiniMax M3, it focuses on ultra-fast Agent response and high-concurrency scenarios while retaining a 1-million-token context window.
The official M3.1-Flash-Preview is offered exclusively through MiniMax's M Plan subscription with thinking always enabled. The iCreat version provides independent pay-as-you-go billing and allows the thinking mode to be disabled to reduce latency, making it suitable for production environments that require flexible cost control and low-latency responses.
Core Capabilities
Ultra-Fast Agent Response
MiniMax M3.1 Flash is optimized for high-throughput, low-latency Agent interactions. When thinking is disabled, the model minimizes reasoning overhead, making it ideal for scenarios requiring immediate feedback and high request concurrency.
Multimodal Understanding
The model accepts text, image (image_url), and video (video_url) inputs within a unified content array structure. This enables comprehensive multimodal understanding for tasks such as visual question answering, video content analysis, and image-based code debugging.
Massive Long-Context Throughput
With a 1-million-token context window, MiniMax M3.1 Flash can ingest extensive documentation, large codebases, and long conversation histories. The max_tokens limit reaches up to 524,288 (512K), allowing for substantial single-pass generation.
Configurable Thinking Mode
The model supports a thinking object with three modes: enabled, adaptive, and disabled. Combined with the reasoning_effort parameter (low, medium, high), developers can precisely balance reasoning depth against response latency based on task complexity.
Pricing
| Token Type | Price (USD) | Unit |
|---|---|---|
| Input | $0.15 | Per million tokens |
| Output | $0.60 | Per million tokens |
| Cache read | $0.015 | Per million tokens |
Note: The above are iCreat platform prices. The official MiniMax M3 (the previous generation formal release in the same series) is priced at $0.30 for input, $1.20 for output, and $0.06 for cache read per million tokens (input prices double when exceeding 512K). The M3.1-Flash-Preview is subscription-only (M Plan) on the official channel with no pay-as-you-go pricing. iCreat offers M3.1 Flash pay-as-you-go billing at $0.15/$0.60, approximately half of the official M3 pay-as-you-go rates. iCreat platform prices shall prevail.
Application Scenarios
- Ultra-fast Agent response and high-concurrency automated customer service
- Code generation, analysis, and rapid bug fixing in development workflows
- Long-document summarization and massive codebase comprehension
- Multimodal tasks involving image and video understanding
- High-throughput data extraction and processing pipelines
Model Comparison
Comparison Table 1: MiniMax M3.1 Flash vs MiniMax M3
| Model | iCreat / Official Pricing | Context Window | Thinking Mode | Target Scenario |
|---|---|---|---|---|
| MiniMax M3.1 Flash | $0.15 / $0.60 / $0.015 (iCreat) | 1 million | Can be disabled to reduce latency | Ultra-fast Agent response, high concurrency |
| MiniMax M3 | $0.30 / $1.20 / $0.06 (Official) | 1 million | Always enabled | General-purpose flagship, open-source |
Note: M3.1 Flash is the pay-as-you-go version of the subscription-based Preview. The iCreat unit price is approximately half of the official M3 pay-as-you-go price.
Comparison Table 2: Cross-Vendor High-Speed Agent Model Comparison
| Model | Input / Output / Cache Read | Context Window | Modality | Note |
|---|---|---|---|---|
| MiniMax M3.1 Flash (MiniMax) | $0.15 / $0.60 / $0.015 | 1 million | Text / Image / Video | Lowest input price among the three |
| DeepSeek V4.1 Flash 1017 (DeepSeek) | $0.3 / $1.2 / $0.006 | 1 million | Text / Image | MIT open-source |
| Gemini 4 Argon (Google) | $2 / $10 (promo price, $4 / $20 from Dec) | 1 million output limit | Text / Image / Video | DeepSWE v1.1 77.9% |
Note: M3.1 Flash has the lowest input price among the three, making it highly suitable for ultra-high concurrency Agent scenarios.
Why Choose MiniMax M3.1 Flash?
- Cost-Effective Pay-As-You-Go: iCreat platform pricing is approximately half of the official M3 pay-as-you-go rates, with no subscription required.
- Ultra-Low Latency: The thinking mode can be disabled to minimize reasoning overhead, ensuring rapid responses for time-sensitive Agent tasks.
- Massive Context: A 1-million-token context window handles extensive documents and large codebases in a single pass.
- Multimodal Support: Processes text, images, and videos within a unified API structure for versatile application development.
- OpenAI Compatibility: Uses the standard OpenAI Chat Completions endpoint, allowing seamless integration with existing SDKs and workflows.
Specifications
| Field | Value |
|---|---|
| Model Name | MiniMax M3.1 Flash |
| Developer | MiniMax |
| Model ID | minimax-m3-1-flash |
| Model Type | Multimodal Large Language Model (High-throughput, Low-latency) |
| Base URL | https://api.icreat.ai/v1 |
| Endpoint | https://api.icreat.ai/v1/chat/completions |
| Authentication | Authorization: Bearer ${ICREAT_API_KEY} |
| Input Modalities | Text, Image (image_url), Video (video_url) |
| Output Modalities | Text |
| Context Window | 1 million tokens |
| max_tokens Limit | Recommended < 128K, Max 524,288 (512K) |
| Thinking Mode | enabled / adaptive / disabled |
| Reasoning Budget | low / medium / high |
| Streaming | Supported (stream: true) |
| Billing Unit | Per million tokens (USD) |
| API Mode | Synchronous POST (JSON / SSE) |
Architecture
MiniMax M3.1 Flash is built as a high-throughput, low-latency multimodal large language model. It adopts a sparse Mixture-of-Experts (MoE) architecture, inheriting the 1-million-token context window from the M3 series while optimizing inference speed and concurrency capabilities. The model integrates a configurable thinking mechanism, allowing the reasoning process to be toggled or adapted based on the thinking object and reasoning_effort parameter. It processes multimodal inputs by parsing a unified content array containing text, image_url, and video_url blocks, converting them into a unified representation for the model backbone. The output is generated as text via the OpenAI-compatible Chat Completions interface, supporting both full JSON responses and SSE streaming.
Notes
- The
modelparameter must use the iCreat model code:minimax-m3-1-flash. - iCreat pricing ($0.15/$0.60/$0.015) is the platform pay-as-you-go rate. Actual settlement is subject to platform billing and the
costUSDfield. max_tokensis recommended to be less than 128K, with a hard limit of 524,288.thinking.typecan be set toenabled,adaptive, ordisabled; setting it todisabledreduces latency.reasoning_effortcan be set tolow,medium, orhigh, and is used in conjunction withthinking.type.- Multimodal content must be passed in the
contentarray by type:text,image_url(containing aurl), andvideo_url(containing aurl). All URLs must use HTTPS. - The context window supports up to 1 million tokens.
- The official channel's M3.1-Flash-Preview is subscription-based (M Plan) with thinking always enabled; the iCreat version provides independent pay-as-you-go billing and allows thinking to be disabled.
Frequently Asked Questions
How do I call MiniMax M3.1 Flash?
You can call it using the OpenAI-compatible endpoint at https://api.icreat.ai/v1/chat/completions. Set the model parameter to minimax-m3-1-flash, use Bearer token authentication with your iCreat API key, and you can enable streaming by setting stream to true.
How is the cost calculated?
The cost is calculated based on token usage: $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cache-read tokens. This is approximately half of the official M3 pay-as-you-go price. The usage field in the response provides the token statistics used for billing.
What input modalities are supported?
The model supports multimodal inputs including text, images, and videos. You provide these by constructing a content array with objects specifying the type (such as text, image_url, or video_url). The output is always text.
How do I configure the thinking mode?
The thinking mode is configured via the thinking object. The thinking.type parameter accepts three values: enabled, adaptive, or disabled. Setting it to disabled minimizes latency. The reasoning_effort parameter (low, medium, high) controls the reasoning budget and is used in conjunction with thinking.type.
What are the context and output limits?
The model supports a context window of 1 million tokens. For output, the max_tokens parameter is recommended to be kept below 128K, with a maximum limit of 524,288 tokens (512K).
What is the difference between this and the official M3.1-Flash-Preview subscription?
The official M3.1-Flash-Preview is subscription-based (M Plan) with no pay-as-you-go pricing and thinking always enabled with five effort levels. The iCreat version provides independent pay-as-you-go billing at $0.15/$0.60 and allows the thinking mode to be disabled to reduce latency.