We Use Cookies

    We use cookies to enhance your browsing experience, analyze site traffic, and personalize content. By clicking "Accept All", you consent to our use of cookies. You can customize your preferences or reject non-essential cookies.

    Learn more about our cookie policy
    Models · EU · Azure

    EU Model Expansion: Claude, DeepSeek V4 Flash, DeepSeek V3.2 & GPT-5.4 — now EU-hosted

    AI·Collab expands its EU-hosted lineup to 10 models — including Anthropic Claude for the first time. All on Azure Sweden Central, all agentic-capable.

    Models · EU
    ≈ 8 min read
    EU-hosted
    ZDR
    AGENTIC

    The short version

    Six new models are now available on Azure Sweden Central — bringing AI·Collab's EU-hosted roster to 10 models from 5 vendors. For the first time, Anthropic Claude (Sonnet 4.6, Opus 4.7, Haiku 4.5) is available EU-hosted on AI·Collab. DeepSeek V4 Flash brings a 1-million-token context window. GPT-5.4 adds Microsoft's latest frontier to the EU lineup. Every new model is agentic-capable and runs on a zero-data-retention path.

    10 EU models
    All agentic
    5 vendors
    Azure Sweden Central

    Why this expansion?

    Since launching EU routing, the ask has been consistent: teams want frontier-tier models and EU data residency without compromise. Until now, EU hosting covered GPT variants, Mistral, Grok, and earlier DeepSeek models. With this update, Claude arrives EU-hosted for the first time — and DeepSeek's 1-million-token model and Microsoft's latest GPT-5.4 join the EU fleet. More EU vendors also means more redundancy: if one model or provider has an issue, strong alternatives with the same EU profile are right there in the picker.

    The 6 new EU-hosted models

    Claude Sonnet 4.6 (EU)

    azure/claude-sonnet-4-6

    Anthropic's strongest all-round model for coding, analysis, and agentic workflows. 200K token context.

    4,255 / 21,272 credits per 1M input / output tokens

    Claude Opus 4.7 (EU)

    azure/claude-opus-4-7

    Maximum quality for complex, long-horizon agentic tasks where reliability matters more than speed. 200K context.

    7,091 / 35,452 credits per 1M input / output tokens

    Claude Haiku 4.5 (EU)

    azure/claude-haiku-4-5

    Fastest and most affordable Claude — ideal for high-volume use, classification, and latency-sensitive applications.

    1,419 / 7,091 credits per 1M input / output tokens

    DeepSeek V4 Flash (EU)

    azure/deepseek-v4-flash

    1-million-token context window, optimised for agentic coding and long-horizon reasoning. Full repositories in one pass.

    199 / 398 credits per 1M input / output tokens

    DeepSeek V3.2 (EU)

    azure/deepseek-v3-2

    Strong reasoning combined with robust tool-calling. Solid choice for structured agentic workflows with external tools.

    358 / 536 credits per 1M input / output tokens

    GPT-5.4 (EU)

    azure/gpt-5.4

    Microsoft's latest frontier model from EU infrastructure — coding, complex reasoning, and professional use cases.

    3,546 / 21,272 credits per 1M input / output tokens

    All 10 EU-hosted models at a glance

    ModelIn Credits / 1MOut Credits / 1MAgentic
    GPT-5 Mini (EU)142567
    GPT-OSS-120B (EU)142142
    Grok-4.1 Fast Reasoning (EU)283708
    Mistral Large 3 (EU)7092,126
    DeepSeek V4 Flash (EU)
    NEW
    199398
    DeepSeek V3.2 (EU)
    NEW
    358536
    Claude Sonnet 4.6 (EU)
    NEW
    4,25521,272
    Claude Opus 4.7 (EU)
    NEW
    7,09135,452
    Claude Haiku 4.5 (EU)
    NEW
    1,4197,091
    GPT-5.4 (EU)
    NEW
    3,54621,272

    All models: Azure Sweden Central · ZDR path · Live prices always in the model catalog.

    What agentic-capable means

    All new models are flagged as AGENTIC in AI·Collab, which unlocks the full tool suite:

    Memory — the model retains context across conversations
    Tool-calling — the model can use external tools: web search, file access, knowledge bases
    Long-horizon tasks — multi-step workflows without manual checkpoints

    EU hosting, redundancy & data sovereignty

    What EU hosting concretely means for your organisation:

    Routing and inference via European infrastructure (Azure Sweden Central)
    No routing through US endpoints for these models
    ZDR-compatible path: no data retention for training purposes
    Geo-diversification: 10 EU models from 5 vendors — Microsoft, Anthropic, DeepSeek, xAI, Mistral

    Hosting location is a meaningful building block for DPIA documentation and procurement — not a substitute for individual legal assessment. Your DPO / counsel decides what is sufficient for your use case.

    How to get started

    Open the model picker at chat.aicollab.app
    Search for the EU model you want — or filter by the EU badge
    Look for EU, ZDR, and AGENTIC labels on the card
    For sensitive tasks: keep web search disabled to avoid extra external calls

    FAQ

    Do I need to do anything to use the new models?

    No. All new models are available directly in the model picker at chat.aicollab.app. Just select and start.

    Are prices different from the non-EU versions of the same models?

    Yes — EU-hosted models carry a slightly higher credit rate than the OpenRouter path for the same models, reflecting the higher infrastructure cost of EU Azure hosting. Exact values are always current in the model catalog.

    What is the difference between Claude Sonnet, Opus, and Haiku?

    Simplified: Haiku = fast and affordable, Sonnet = balanced for professional daily use, Opus = maximum quality for complex long-horizon tasks. All three are now EU-hosted.

    Can I combine EU models with Memory and web search?

    Memory: yes, all agentic models support Memory. Web search: use it deliberately — external web search calls may go through external endpoints. For maximum EU data-path control, keep web search disabled for sensitive tasks.

    Why was DeepSeek-R1 replaced by V4 Flash?

    DeepSeek V4 Flash offers a significantly larger context window (1M vs. 64K tokens), is optimised for agentic coding, and has a lower credit rate for comparable reasoning. R1 remains available via other paths.

    Figures and badges reflect the product configuration at publish time. The model catalog is always the authoritative source for live prices and labels.

    Related Articles

    Ready to Experience 300+ AI Models?

    Get started today. Access models from OpenAI, Google, Anthropic, Grok and more.

    GDPR compliant · Zero data retention · Cancel anytime