--- title: "DeepSeek V4 Flash API hosted in the EU" description: "Run DeepSeek V4 Flash (0731) on inference.eu: a 1M-token open reasoning model behind an OpenAI-compatible API, processed in EU data centres, not in China." canonical: "https://inference.eu/models/deepseek-v4-flash/" language: "en" --- Reasoning · DeepSeek # Run DeepSeek V4 Flash in the EU through an OpenAI-compatible API DeepSeek V4 Flash is DeepSeek's efficient mixture-of-experts reasoning model with a one-million-token context, released under the MIT licence. This page covers the official 0731 release, which replaced the April preview. On inference.eu your requests are processed in EU data centres, never in China, and never used for training. [Contact sales](https://inference.eu/contact/)[Example request](#example) 1. 01284B · 13B 2. 02EU data centres 3. 03MIT ## What DeepSeek V4 Flash is DeepSeek V4 Flash is the smaller member of the DeepSeek V4 series. It is a mixture-of-experts model with 284 billion parameters, 13 billion of them active per token, and a context window of one million tokens. DeepSeek published a preview in April 2026 and the official release, DeepSeek-V4-Flash-0731, at the end of July 2026; we treat both as one model and serve the 0731 version. The architecture combines two compressed sparse attention schemes to keep very long contexts affordable. The model supports three reasoning levels through `reasoning_effort`: low, high and max. DeepSeek describes the 0731 release as having substantially stronger agentic abilities than the preview. It reads and writes text. ## What it does well, and where it falls short DeepSeek V4 Flash is built for long-horizon agent work and coding at a low active-parameter cost. It can hold an entire repository, a long case file or months of chat history in one request, and at high effort it plans and checks its own work over many steps. For a model with 13 billion active parameters, it is unusually capable at tool-using tasks. It is text-only; for images use DeepSeek V4.1 Flash. The max reasoning level can generate very long outputs, and DeepSeek recommends allowing up to 384K output tokens there, so cost and time grow quickly. For simple chat, low effort or a smaller model is the better choice. Its training reflects its maker's choices, including on politically sensitive topics. ## Calling DeepSeek V4 Flash through the API inference.eu uses the OpenAI request format, so the OpenAI SDKs, LangChain, LlamaIndex and most coding agents work as they are. Set the base URL to the inference.eu API, use your inference.eu key and set the model to `deepseek-v4-flash`. DeepSeek ships its own prompt encoding instead of a standard chat template; we handle it, so you send normal messages. Set `reasoning_effort` to `low`, `high` or `max`. DeepSeek recommends `temperature` 1.0 with `top_p` 0.95 for agent tasks. Tool calls and streaming work as usual, and the reasoning is returned separately from the answer. The Anthropic Messages format is available too. ## DeepSeek without sending data to China Many European companies avoid DeepSeek's own app and API because data would be processed in China. The open weights remove that problem. On inference.eu, DeepSeek V4 Flash runs on infrastructure in the European Union: your prompts and responses are processed in EU data centres, nothing goes to DeepSeek, and nothing is used to train any model. Every customer receives a data processing agreement under Article 28 of the GDPR and our subprocessor list, so you can show that no third-country transfer takes place for inference. Billing is in euros. The MIT licence allows commercial use with almost no conditions. ## DeepSeek V4 Flash Maker DeepSeek Made in China Type Reasoning Parameters 284B Active parameters 13B Context window 1,048,576 tokens Licence [MIT](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/main/LICENSE) Released July 2026 Model ID deepseek-v4-flash Processing EU data centres Training on your data Never Source [huggingface.co](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731) deepseek-v4-flash · 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="deepseek-v4-flash", messages=[{"role": "user", "content": "Summarise this contract in five bullet points."}], ) print(response.choices[0].message.content) ``` ## What teams use it for - 1 ### Coding agents on large repositories Let an agent read a whole codebase, plan a change across many files and verify it, using the one-million-token context. - 2 ### Long case and contract review Analyse a complete case file or a full set of contracts in a single request and reason about inconsistencies. - 3 ### Multi-step research tasks Run agents that search, read, compare and summarise sources over many tool calls before producing a report. ## DeepSeek V4 Flash: common questions The questions buyers, developers and legal teams ask us most. Something missing? Ask us directly. Data safety depends on where the model runs. On inference.eu, DeepSeek V4 Flash runs on EU infrastructure from openly published weights: no data goes to DeepSeek or to China. What remains is the model's own behaviour, which reflects its training, including on politically sensitive topics. Test it on your own prompts. Compliance depends on your processing. On inference.eu, 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. That avoids the transfer to China that makes DeepSeek's own service hard to justify. DeepSeek V4 Flash supports one million tokens of context. That is enough for a large codebase, long legal files or very long conversations. Reasoning output also needs room, so leave space when you use high or max effort. DeepSeek released a preview of V4 Flash in April 2026 and the official DeepSeek-V4-Flash-0731 at the end of July. The official release keeps the same size and context but has substantially better agentic abilities. On inference.eu, \`deepseek-v4-flash\` refers to the 0731 version. Yes. The repository and the model weights are licensed under the MIT licence, which allows commercial use, modification and redistribution. You only need to keep the copyright and licence notice if you redistribute the weights. Pass \`reasoning\_effort\` with \`low\`, \`high\` or \`max\`. Low gives quick answers, high adds careful step-by-step reasoning, and max pushes reasoning as far as it goes. DeepSeek recommends a very large output budget for high and max. No, V4 Flash is text-only. DeepSeek V4.1 Flash, released in September 2026, adds native image input. For document scans, DeepSeek-OCR 2 converts pages into Markdown that V4 Flash can then read. V4.1 Flash is a newer multimodal model with a different architecture, image input and a continuous reasoning-effort setting from 1 to 100. V4 Flash is text-only with three effort levels. Both have a one-million-token context and the MIT licence. Change your OpenAI client's base URL to the inference.eu API, use your inference.eu key and set the model to \`deepseek-v4-flash\`. Messages, tools and streaming keep working. Raise \`max\_tokens\` if you use high reasoning effort. No. DeepSeek published the weights but is not involved when you call the model on inference.eu. Requests are processed in EU data centres to produce your response and are never used to train DeepSeek V4 Flash or any other model. [Ask a question](https://inference.eu/contact/) ## Related [All models](https://inference.eu/models/#directory) - [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/) - [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/) - [gpt-oss-120bRun OpenAI's open-weight gpt-oss-120b reasoning model on inference.eu: the OpenAI API format, with requests processed in EU data centres, billed in euros.Reasoning](https://inference.eu/models/gpt-oss-120b/) - [Qwen3 235B A22B InstructRun Qwen3 235B A22B Instruct (2507) on inference.eu: a 256K-context open model behind an OpenAI-compatible API, processed in EU data centres.Chat](https://inference.eu/models/qwen3-235b-a22b-instruct/) ## 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