MiniMax-M3 is MiniMax's flagship open-weight foundation model and the first to combine three frontier capabilities at once: frontier-level coding and agentic performance, a 1M-token context window, and native multimodality. It accepts text, image, and video inputs with text output, and is powered by the proprietary MiniMax Sparse Attention (MSA) architecture, which sustains up to 1M tokens of context (with a guaranteed minimum of 512K) - the foundation for long-range agent tasks, long-horizon coding, and long-video understanding. Multimodality is a native core capability rather than an add-on: the data pipeline was rebuilt to scale pretraining to 100T+ tokens with multimodal training from step zero, deeply aligning textual and visual semantic spaces. M3 achieves top-tier results across coding and agentic benchmarks spanning software engineering, terminal execution, and autonomous browsing (scoring 83.5 on BrowseComp), with autonomous task decomposition, tool invocation, and multi-step reasoning. It is well suited to AI coding assistants, automated workflows, and long-running asynchronous agent pipelines where coherence over extended sessions matters.
MiniMax M3 is a large language model developed by Minimax, accessible through OrcaRouter's OpenAI-compatible API. It accepts text, image, and video inputs, and has a context window of 1,048,576…
MiniMax M3 can process long documents up to 1,048,576 tokens in a single pass. This allows it to perform tasks such as summarization, question answering, and information extraction on entire research papers, legal briefs, or technical manuals without losing context. The model can also handle multi-document inputs, such as a collection of articles, and generate a coherent output. For applications requiring very long outputs, the 512,000-token generation limit enables production of full-length reports or code. This capability is particularly useful for enterprise document processing and data analysis workflows.
MiniMax M3 accepts image and video inputs alongside text, allowing it to reason about visual content. Images and videos are tokenized and embedded into the same context as text. The model can answer questions about the contents of an image, describe a video scene, or perform optical character recognition. For video, it can process multiple frames or the entire video as long as the combined token count stays within the 1,048,576 token window. This multimodal support makes it suitable for tasks such as captioning, visual question answering, and video analysis, all accessible through the OrcaRouter API.
MiniMax M3 is optimized for tasks that require a very large context window (1,048,576 tokens) or multimodal input (text, image, video). If your application involves processing long documents, entire books, or hours of video in a single API call, M3's context size is a strong advantage. It also suits scenarios where you need to generate up to 512,000 tokens of output without multiple requests. For simpler tasks—such as short-form text generation, summarization of a few paragraphs, or single-image descriptions—a cheaper model with a smaller context window may be more cost-effective. Evaluate your average input and output lengths to decide.
MiniMax M3 is best suited for tasks that leverage its large context window and multimodal capabilities. Examples include: analyzing entire research papers or legal documents with question-answering; extracting structured data from long multi-page forms; generating extensive reports or code from a large set of inputs; video content analysis, such as summarizing a lecture or identifying objects across many frames; and building applications that require storing and reasoning over a large history of user interactions within a single conversation. Its pricing—$0.30 per million input tokens and $1.20 per million output tokens—makes it economical for high-volume, long-context workflows.
MiniMax M3 achieved a score of 83.5 on the BrowseComp benchmark. BrowseComp evaluates a model's ability to navigate web pages and extract relevant information from them. This test typically involves simulating a user who browses a series of web pages and then answers questions based on the content. A higher score indicates better performance in understanding web layouts, following links, and synthesizing information from multiple pages. The 83.5 score places MiniMax M3 in a competitive position among models tested on this benchmark, particularly for tasks involving web-based information retrieval and reasoning.
The primary strength of MiniMax M3 shown by its 83.5 BrowseComp score is its ability to handle web-based information-seeking tasks. This indicates strong reading comprehension and navigation capabilities. However, the model's performance on other common benchmarks (such as MMLU, HumanEval, or GSM8K) has not been publicly disclosed as part of the provided facts. Therefore, its relative performance in general knowledge, code generation, or math reasoning is not quantified here. Users should evaluate the model on their own tasks. The large context window and multimodal support are additional strengths not captured by BrowseComp alone.
Inference speed and latency for MiniMax M3 depend on several factors including input length, output length, and server load. No specific latency figures are provided in the available facts. Generally, models with large context windows may have higher latency for very long inputs due to the computational cost of processing many tokens. The output limit of 512,000 tokens means that generating very long responses will take proportionally longer. When using OrcaRouter's API, latency will be similar to other large models of comparable size. For near-real-time applications requiring low latency, consider models with smaller context windows.
MiniMax M3 is priced at $0.30 per 1 million input tokens and $1.20 per 1 million output tokens. These rates reflect the provider's pricing with zero markup added by OrcaRouter. Input tokens include all tokens in the prompt, including system messages, user messages, and any image or video tokens. Output tokens are the tokens generated in the response. Cost scales linearly with token usage. For example, a request with 100,000 input tokens and 10,000 output tokens would cost $0.03 for input and $0.012 for output, totaling $0.042. There are no additional platform fees or minimum commitments.
The provided facts do not mention any specific discount programs or caching benefits for MiniMax M3 on OrcaRouter. Pricing is straightforward per token at the rates above. OrcaRouter does not offer automatic caching or reduced rates for repeated inputs based on available information. Users should consult OrcaRouter's documentation or contact their support team for the most up-to-date details on any potential cost-saving features. In the absence of such programs, the cost is directly proportional to the number of input and output tokens consumed.
MiniMax M3's pricing of $0.30 per million input tokens and $1.20 per million output tokens is competitive for a model offering a 1,048,576 token context window and multimodal input. Without specific competitor pricing provided in the facts, a general comparison: many large models with smaller context windows (e.g., 128k or 200k tokens) are priced similarly per token, but their context limit requires multiple API calls for very long documents. MiniMax M3's larger context may reduce total cost for tasks that would otherwise need chunking and multiple requests. Users should calculate total token usage across their typical workflow to compare.
To use MiniMax M3 through OrcaRouter, send requests to the base URL https://api.orcarouter.ai/v1. The model identifier is "minimax/minimax-m3". Authentication is done by passing your OrcaRouter API key in the Authorization header as "Bearer YOUR_API_KEY". The API is compatible with the OpenAI SDK, so you can use the same client library by updating the base URL and API key. For example, in Python with the openai package, set `client = OpenAI(base_url="https://api.orcarouter.ai/v1", api_key=YOUR_API_KEY)` and then call `client.chat.completions.create(model="minimax/minimax-m3", messages=[...])`.
When calling MiniMax M3 through OrcaRouter's OpenAI-compatible API, you can use standard parameters such as `messages`, `max_tokens`, `temperature`, `top_p`, `frequency_penalty`, `presence_penalty`, `stop`, and `stream`. The `max_tokens` parameter should not exceed the model's maximum output of 512,000 tokens. Input tokens (in `messages`) plus output tokens must stay within the context window of 1,048,576 tokens. For multimodal inputs, include images or video in the `content` array using the appropriate format (e.g., `image_url` for images). Consult the OpenAI API documentation for full parameter descriptions.
To migrate an existing application that uses OpenAI's API to MiniMax M3 on OrcaRouter, you need to change two things: the base URL and the model ID. Replace your OpenAI base URL with "https://api.orcarouter.ai/v1". Change the model string to "minimax/minimax-m3". Also update your API key to an OrcaRouter API key. The message format remains identical—OpenAI-compatible. No other code changes are required. If your application uses streaming, the stream interface is also compatible. Test with a small request first to ensure connectivity and that the model responds as expected.
Authentication to OrcaRouter's API is done via an API key. You must include your OrcaRouter API key in the Authorization header of each request as "Bearer YOUR_API_KEY". API keys are issued by OrcaRouter; you can obtain one by signing up on their platform. Keep your key secure and do not expose it in client-side code. The API supports key-based access only; no OAuth or other authentication methods are documented for this endpoint. If you are using the OpenAI Python library, set the `api_key` parameter to your OrcaRouter key when initializing the client.
MiniMax M3 offers a context window of 1,048,576 tokens, which is among the largest available. Many competing long-context models offer 128K, 200K, or 1M tokens, but few also support image and video input. The 83.5 BrowseComp score suggests strong performance on web-based information retrieval tasks. However, without other benchmark scores, a comprehensive comparison is limited. In terms of pricing, at $0.30 input / $1.20 output per million tokens, it is moderately priced for its context size. Users should compare total cost for their typical input lengths and output generation needs.
Choose MiniMax M3 when your application requires a very large context window (up to 1M tokens) or multimodal input (text, image, video). If you need to process a long document or video in a single API call without chunking, M3's context size is a key advantage. It is also suitable if you need to generate up to 512,000 tokens of output. For tasks that are purely text-based with smaller context requirements (e.g., under 100K tokens), a cheaper model with a smaller context window may be more cost-effective. Additionally, if you need the fastest inference speed, models with smaller context windows may respond quicker.
Data privacy for MiniMax M3, when accessed through OrcaRouter, is governed by OrcaRouter's data handling policies. OrcaRouter does not add a markup to the provider's pricing, but the data processing infrastructure is managed by OrcaRouter. The model provider (Minimax) may also have its own data usage terms. Users should review both OrcaRouter's privacy policy and Minimax's terms regarding data retention, training, and encryption. No specific security certifications or data residency options are provided in the available facts. For sensitive data, consider contacting OrcaRouter directly for details on data handling and compliance.
OpenAI-compatible — keep the SDK you already use
https://api.orcarouter.ai/v1from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key="$ORCAROUTER_API_KEY",
)
response = client.chat.completions.create(
model="minimax/minimax-m3",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)include_reasoningmax_tokensreasoningresponse_formattemperaturetool_choicetoolstop_p| Input / 1M tokens | $0.300 |
| Output / 1M tokens | $1.20 |
| Cache read / 1M | $0.060 |
| Currency | USD |
Estimate based on list price
Estimate only — actual token counts depend on the provider's tokenizer.
GET /api/public/models/minimax/minimax-m3Open @misc{orcarouter_minimax_m3,
title = {MiniMax M3 API},
author = {MiniMax},
year = {2026},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/minimax/minimax-m3}
}MiniMax. (2026). MiniMax M3 API. OrcaRouter. https://www.orcarouter.ai/models/minimax/minimax-m3