Power BI 19 May 2026 6 min read

Claude in Power BI: How to Use Anthropic's AI Models in Your Fabric Workspace

If your Power BI team is running everything on GPT, you are leaving capability on the table. Claude — Anthropic's family of large language models — offers distinct advantages for specific Power BI workloads: longer context windows, better document analysis, more nuanced reasoning, and cleaner structured output.

If your Power BI team is running everything on GPT, you are leaving capability on the table. Claude — Anthropic's family of large language models — offers distinct advantages for specific Power BI workloads: longer context windows, better document analysis, more nuanced reasoning, and cleaner structured output.

The good news: Claude is available directly inside Microsoft Foundry, which means it integrates natively with Fabric AI Functions, SynapseML, and your Power BI workspace. No third-party connectors. No separate subscriptions. Just deploy and use.

This guide covers how to set up Claude in your Power BI environment, when to choose Claude over GPT, and what each model is best at.

What Claude Models Are Available in Foundry?

As of May 2026, Microsoft Foundry offers the following Claude models via the Azure Marketplace:

Model Context Window Output Limit Best For
Claude Opus 4.7 1,000,000 tokens 128,000 tokens Complex reasoning, long documents, code generation
Claude Sonnet 4.6 1,000,000 tokens 128,000 tokens Balanced performance and cost, general-purpose
Claude Haiku 4.5 200,000 tokens 64,000 tokens Fast, cheap tasks — classification, extraction, simple summaries

All three models support text, image, and code input. All support tool calling (file search and code execution). All support multiple languages: English, French, Arabic, Chinese, Japanese, Korean, Spanish, and Hindi.

When to Choose Claude Over GPT

Claude and GPT are both excellent models, but they have different strengths. Here is when Claude is the better choice for Power BI workloads:

Claude wins at:

GPT wins at:

The practical approach: Use GPT-4.1-mini (the default) for simple, high-volume tasks like sentiment analysis and grammar correction. Use Claude Sonnet 4.6 for summarisation, classification, and extraction where quality matters. Use Claude Opus 4.7 for complex reasoning, long documents, and code generation.

How to Deploy Claude in Microsoft Foundry

Step 1: Create a Foundry Resource

  1. Go to ai.azure.com
  2. Sign in with your Azure account (must have an active subscription)
  3. Click Create new → Foundry resource
  4. Select your subscription, resource group, and region
  5. Name your resource (e.g., my-fabric-ai)
  6. Click Review + Create → Create

Note: Claude models require a paid Azure subscription. They are not supported on free trials, student accounts, startup credits, or Enterprise Agreement subscriptions in South Korea.

Step 2: Deploy a Claude Model

  1. In your Foundry resource, navigate to Model catalog
  2. Search for "Claude"
  3. Select Claude Sonnet 4.6 (recommended starting point)
  4. Click Deploy → name your deployment (e.g., claude-sonnet)
  5. Note the endpoint URL and API key — you will need these

Step 3: Configure AI Functions to Use Claude

By default, Fabric AI Functions use GPT-4.1-mini. To switch to Claude:

  1. Open your Fabric workspace
  2. Navigate to Settings → AI Functions configuration
  3. Change the model provider from Default (Azure OpenAI) to Microsoft Foundry
  4. Enter your Foundry resource endpoint and API key
  5. Select your Claude deployment (e.g., claude-sonnet)
  6. Save the configuration

Now all 9 AI Functions will use Claude instead of GPT:


      # Before (default GPT-4.1-mini)
      df['summary'] = ai.summarize(df['review_text'])
      
      # After (now uses Claude Sonnet 4.6)
      df['summary'] = ai.summarize(df['review_text'])
      # Same code, different model — the configuration change is global

Step 4: Use Claude via SynapseML for Advanced Scenarios

For more control — custom prompts, batch processing, specific model parameters — use SynapseML directly:


      from synapse.ml.cognitive.openai import OpenAIChatCompletion
      
      # Configure to point at your Claude deployment
      chat = OpenAIChatCompletion(
          deploymentName="claude-sonnet",
          endpoint="https://your-foundry-resource.cognitiveservices.azure.com/",
          apiKey="your-api-key",
          userPrompt="Summarize this customer feedback: {text}",
          temperature=0.3
      )
      
      # Process at scale across your Spark cluster
      results = chat.transform(spark_df)

SynapseML distributes the processing across your Fabric Spark cluster, handling batching, concurrency, and error recovery automatically.

Real-World Use Cases

Contract Analysis (Claude Opus 4.7)

Upload entire contracts (up to 1M tokens) and extract key terms, obligations, and risk clauses. Claude's long context window means you do not need to chunk documents.


      df['contract_analysis'] = ai.generate_response(
          df['contract_text'],
          instructions="Extract: 1) Contract duration 2) Payment terms 3) Termination clauses 4) Liability caps. Format as JSON.",
          response_format='json'
      )

Customer Feedback Processing (Claude Sonnet 4.6)

Process thousands of customer reviews with nuanced sentiment analysis and theme extraction. Claude's summarisation produces more balanced, less hallucination-prone results than smaller models.


      df['sentiment'] = ai.analyze_sentiment(df['feedback'])
      df['themes'] = ai.extract(df['feedback'], labels=["complaint", "praise", "suggestion", "question"])
      df['summary'] = ai.summarize(df['feedback'], instructions="Focus on actionable insights for the product team")

DAX Code Generation (Claude Opus 4.7)

Generate complex DAX measures from natural language descriptions. Claude's code generation is particularly strong for DAX, which requires understanding of filter context and table relationships.


      df['dax_measure'] = ai.generate_response(
          df['measure_description'],
          instructions="Convert this description to a DAX measure. Use proper CALCULATE, FILTER, and ALL functions. Return only the DAX code."
      )

Executive Report Summarisation (Claude Sonnet 4.6)

Generate executive summaries from detailed Power BI reports. Claude produces more nuanced, balanced summaries that acknowledge limitations and uncertainties.


      df['exec_summary'] = ai.summarize(
          df['report_content'],
          instructions="Write a 200-word executive summary. Highlight key trends, risks, and recommendations. Use formal business language."
      )

Pricing

Claude models are billed through Azure Marketplace as SaaS offerings. Pricing varies by model and region.

Approximate pricing (May 2026):

For comparison, GPT-4.1-mini (the default for AI Functions) costs approximately $0.40 per 1M input tokens.

Cost management tips:

Restrictions

Claude models in Foundry have some restrictions to be aware of:

Summary

Claude brings real capability to Power BI that GPT alone cannot match. The 1M context window alone is a game-changer for document analysis and long-form summarisation. With Foundry integration, switching your AI Functions from GPT to Claude is a configuration change — no code changes required.

Start with Claude Sonnet 4.6 for your next summarisation or extraction task. Compare the output quality against GPT-4.1-mini. If the difference justifies the cost, consider Claude Opus 4.7 for your most complex analytical workloads.

The models are deployed. The integration is built. The only question is whether your Power BI team is using them.

Related Articles

P

PowerBI.ai

Expert insights on Power BI, Microsoft Fabric, and AI-powered business intelligence.