--- title: "MiniMax M3 API hosted in the EU" description: "Run MiniMax M3 on inference.eu: an open multimodal model with 1M context for coding and agent work, OpenAI-compatible and processed in EU data centres." canonical: "https://inference.eu/models/minimax-m3/" language: "en" --- Reasoning · MiniMax # Run MiniMax M3 in the EU through an OpenAI-compatible API MiniMax M3 is a natively multimodal open-weight model from MiniMax with about 428 billion parameters, 23 billion active, and a one-million-token context. It targets coding and agentic work and reads images and video. On inference.eu you call it with the OpenAI or Anthropic request format, processed in EU data centres and never used for training. [Contact sales](https://inference.eu/contact/)[Example request](#example) 1. 01428B · 23B 2. 02EU data centres 3. 03MiniMax Community License ## What MiniMax M3 is MiniMax M3 was released by the Shanghai-based company MiniMax on 1 June 2026. It is a mixture-of-experts model with roughly 428 billion parameters, of which about 23 billion are active per token. It was trained on mixed modalities from the first step, so text, images and video are part of one model rather than bolted on. Its context reaches one million tokens thanks to MiniMax Sparse Attention, which MiniMax says gives 9 times faster prefill and 15 times faster decoding than M2 at that length. MiniMax describes M3 as the first open-weight model to combine frontier coding, million-token context and native multimodality, and says it can also operate a desktop computer. ## What it does well, and where it falls short M3 is built for long agent tasks: coding, research, office-style work across documents and screens, and computer use where the model reads screenshots and decides on the next click. With about 23 billion active parameters it is moderately priced to run for its capability, and its adaptive reasoning mode decides by itself when extra thinking is worth it. Its licence is more restrictive than MIT or Apache: commercial use is allowed but comes with an attribution requirement and a list of prohibited uses, including military purposes. On the hardest long-horizon coding tasks, Kimi K3 or GLM-5.3 can go further. Like other Chinese models, its training reflects its maker's choices on sensitive topics. ## Calling MiniMax M3 through the API inference.eu supports the OpenAI and the Anthropic request formats. Set the base URL to the inference.eu API, use your inference.eu key and set the model to `minimax-m3`. Coding agents and frameworks that accept a custom endpoint work without further changes. M3 has three reasoning modes through the `thinking` parameter: `enabled` always reasons, `adaptive` lets the model decide, and `disabled` skips reasoning for lower latency and higher throughput. MiniMax recommends `temperature` 1.0 and `top_p` 0.95. Send images as `image_url` content parts next to your text. Tool calls and streaming work as usual. ## Data residency and licence for MiniMax M3 On inference.eu, MiniMax M3 runs on infrastructure in the European Union. Prompts, images, video frames and the model's responses are processed in EU data centres, are not sent to MiniMax in China, and are never used to train any model. Every customer gets a data processing agreement under Article 28 of the GDPR and our subprocessor list. Billing is in euros. Under the MiniMax Community License, if you use M3 in a commercial product you must display "Built with MiniMax M3" on a related website, interface or documentation and send MiniMax a one-time notice; above 20 million US dollars of yearly revenue you need its written authorisation. ## MiniMax M3 Maker MiniMax Made in China Type Reasoning Parameters 428B Active parameters 23B Context window 1,048,576 tokens Licence [MiniMax Community License](https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE) Released June 2026 Model ID minimax-m3 Processing EU data centres Training on your data Never Source [huggingface.co](https://huggingface.co/MiniMaxAI/MiniMax-M3) minimax-m3 · Python ``` import os from openai import OpenAI client = OpenAI(base_url="https://api.inference.eu/v1", api_key=os.environ["INFERENCE_EU_API_KEY"]) response = client.chat.completions.create( model="minimax-m3", messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}], ) print(response.choices[0].message.content) ``` ## What teams use it for - 1 ### Computer-use agents Let an agent read screenshots of desktop or web applications and carry out multi-step tasks on them. - 2 ### Coding with long context Work on large codebases in the one-million-token window, with images of designs or errors as extra input. - 3 ### Video and document review Summarise recorded walkthroughs, slide decks and long documents together in one request. ## MiniMax M3: common questions The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly. Compliance depends on how you process data. On inference.eu, MiniMax M3 requests are processed in EU data centres, not used for training and covered by a data processing agreement under Article 28 GDPR with a published subprocessor list. Nothing is sent to MiniMax or to China. Yes, with conditions. The MiniMax Community License requires commercial users to display Built with MiniMax M3 in a related website, interface or documentation and to send MiniMax a one-time notice. Above 20 million US dollars of yearly revenue you need prior written authorisation. Some uses, such as military purposes, are prohibited. MiniMax M3 supports a context window of one million tokens. Its MiniMax Sparse Attention keeps long inputs affordable compared with earlier MiniMax models. Images and video frames count towards the context. Yes. M3 was trained on text, images and video from the start, so it understands all three natively. It answers in text. MiniMax also describes it as able to operate a desktop computer by reading screenshots. Use the \`thinking\` parameter. Set \`enabled\` to always reason, \`adaptive\` to let the model decide when extra reasoning helps, or \`disabled\` to answer directly with lower latency. Adaptive is a sensible default for mixed workloads. On inference.eu the open weights run in EU data centres, so prompts and media do not leave the EU for inference. The model's behaviour still reflects its training, including on politically sensitive topics. Test it on your own tasks before relying on it. Both are efficient multimodal mixture-of-experts models with one-million-token context. MiniMax M3 is larger and adds video and computer use. GLM-5.3-Flash uses the simpler MIT licence. If licence terms matter most, GLM-5.3-Flash is easier; for screen-driven agents, M3 is worth testing. Kimi K3 is much larger, with 2.8 trillion parameters and 104 billion active, and goes further on the hardest tasks. MiniMax M3 has about 23 billion active parameters, so it is far lighter per token while still offering one million tokens of context and video input. Point your OpenAI client's base URL at the inference.eu API, use your inference.eu key and set the model to \`minimax-m3\`. Messages, images, tools and streaming keep working. Add the \`thinking\` parameter if you want to control reasoning. No. MiniMax published the weights but takes no part in serving your requests on inference.eu. Your prompts and media are processed in EU data centres, used only for your response and never used to train MiniMax M3 or any other model. [Ask a question](https://inference.eu/contact/) ## Related [All models](https://inference.eu/models/#directory) - [GLM-5.3-FlashRun Z.ai's GLM-5.3-Flash on inference.eu: a multimodal open model with 18B active parameters and 1M context, OpenAI-compatible, processed in the EU.Reasoning](https://inference.eu/models/glm-5-3-flash/) - [Kimi K2.6Run Moonshot AI's Kimi K2.6 on inference.eu: a 1-trillion-parameter open agentic model with image input and 256K context, processed in EU data centres.Reasoning](https://inference.eu/models/kimi-k2-6/) - [DeepSeek V4.1 Flash (1017)Run DeepSeek V4.1 Flash (1017) on inference.eu: a multimodal open reasoning model with 1M context, OpenAI-compatible, processed in the EU, not in China.Reasoning](https://inference.eu/models/deepseek-v4-1-flash-1017/) - [Kimi K3Run Kimi K3, Moonshot AI's 2.8-trillion-parameter open-weight model with 1M context and image and video input, on inference.eu with processing in the EU.Reasoning](https://inference.eu/models/kimi-k3/) ## Move your AI workloads to EU infrastructure Tell us which models you use and roughly how much traffic you have. We reply within one working day with pricing and next steps. - Reply within one working day - Pricing for your volume - DPA ready to sign