feat: use inference profiles for on-demand Claude models []

[Mitch Goudy (mgoudy91)]

Summary

Followup to #10706
Fixes Bedrock ValidationException: "Invocation of model ID ... with on-demand throughput isn't supported. Retry your request with the ID or ARN of an inference profile that contains this model."

See:

• https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-support.html
• https://docs.aws.amazon.com/bedrock/latest/userguide/inference-profiles-use.html

Changes

• Inference profile support: Models that require it (Sonnet 4.6, 4.5, 4, 3.5 Haiku, 3 Haiku, v3 Sonnet) now use inference profile IDs (e.g. us.anthropic.claude-sonnet-4-5-20250929-v1:0, global.anthropic.claude-sonnet-4-6) when calling InvokeModel / InvokeModelWithResponseStream.
• Backward compatible: Stored config still uses foundation model id; only the value sent at invoke time changes. v2.1, Instant, Llama, Mistral continue to use foundation model ID.
• Config screen: Models with getInvokeId are no longer marked NOT_IN_REGION solely for missing from ListFoundationModels; availability is determined by the invoke check with the profile ID.

Testing

• npm run build and affected unit tests (featuredModels, Model, aiApi) pass.
• Verified locally: Sonnet 4.6, 4.5, 4, 3.5 Haiku show as available when account has access; 403s are IAM/Marketplace subscription issues, not request format.

[Copilot]

Pull request overview

This PR updates the Bedrock content generator to invoke certain on-demand Anthropic Claude models via inference profile IDs (instead of raw foundation model IDs) to avoid Bedrock ValidationException errors and improve model availability detection.

Changes:

• Add inference profile routing (getInvokeId, region→prefix mapping) for Claude models that require it.
• Pass region into invoke command construction so the correct invoke-time modelId is used for streaming and availability checks.
• Adjust the config UI’s region-availability logic and update related unit test expectations.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
apps/bedrock-content-generator/src/utils/aiApi/index.ts	Pass region into invoke inputs; invoke using computed modelId for streaming and availability checks.
apps/bedrock-content-generator/src/configs/aws/featuredModels.ts	Introduce getInvokeId + inference profile selection helpers; update featured Claude models to use inference profiles.
apps/bedrock-content-generator/src/components/config/model/Model.tsx	Loosen NOT_IN_REGION determination for models that use inference profiles so they still get an invoke-based availability check.
apps/bedrock-content-generator/src/components/config/model/Model.spec.tsx	Update assertion for the NOT_IN_REGION warning text now that fewer models are classified as region-missing from the foundation list.

Comments suppressed due to low confidence (2)

apps/bedrock-content-generator/src/utils/aiApi/index.ts:92

• Now that more models rely on the invoke-time ID (inference profile), getModelAvailability should classify the common “not available in this region / invalid modelId for region” failures as NOT_IN_REGION instead of returning a raw Error (which becomes OTHER_ERROR). Consider handling ValidationException / ResourceNotFoundException (and/or matching their messages) to return NOT_IN_REGION so the config UI shows the correct availability state.

      const invokeInput = model.invokeCommand('', '', 1, this.region);
      await this.bedrockRuntimeClient.send(new InvokeModelCommand(invokeInput));
    } catch (e: unknown) {
      if (!(e instanceof Error)) {
        return Error('An unexpected error has occurred');
      }
      if (e instanceof AccessDeniedException) {
        if (e.message.includes('is not authorized to perform: bedrock:InvokeModel'))
          return 'FORBIDDEN';
        if (e.message.includes("You don't have access to the model with the specified model ID."))
          return 'NOT_IN_ACCOUNT';
      }

apps/bedrock-content-generator/src/components/config/model/Model.tsx:86

• isInRegion is now computed as isInFoundationList || !!featuredModel.getInvokeId, which means models with getInvokeId are treated as “in region” even when they’re absent from ListFoundationModels. This makes the variable name/meaning misleading and can mask genuine NOT_IN_REGION cases unless the invoke error is mapped accordingly. Consider renaming this flag to something like shouldCheckAvailability (or similar) to reflect its purpose and avoid confusion.

          const isInFoundationList = allModels.some((m) => m.modelId === featuredModel.id);
          // Models that use inference profiles may not appear in ListFoundationModels; still run availability check (invoke with profile ID).
          const isInRegion = isInFoundationList || !!featuredModel.getInvokeId;

          return {
            ...featuredModel,
            invokeCommand: featuredModel.invokeCommand,
            availability: isInRegion ? 'AVAILABLE' : 'NOT_IN_REGION',

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.