User-friendly model names?

[slawekp]

Describe your question

My model names are coming from inference profiles, eg.:

arn:aws:bedrock:eu-west-1:123456789000:application-inference-profile/abc1defg2hij

Then in Settings → Model definitions, I define this model with a user-friendly name of AWS Sonnet 5.

When listing Observations of type generation and looking at the Model column, I was expecting to see my user-friendly model name and yet I can only see the inference profile name.

First thing I checked was if the model properly matches the definition and I confirm that it does, because:

• Model ID column matches the model definition ID
• Total cost is being calculated based on my settings
• Observations are listed in Settings → Model definitions → AWS Sonnet 5 → Model observations

What am I missing? Why is the Model column not showing the user-friendly name of the model? Maybe it was never supposed to?

Langfuse Cloud or Self-Hosted?

Self-Hosted

If Self-Hosted

3.224.1

If Langfuse Cloud

No response

SDK and integration versions

No response

Pre-Submission Checklist

• I have checked for existing issues/discussions and consulted Langfuse AI.

[inkeep-ai-assistant[bot]]

You're not missing anything — the Model column shows the raw model string ingested from the trace, not the name of the matched model definition. Since your Bedrock calls report the application inference profile ARN as the model string, that ARN is what appears in the column, even though the definition matched correctly (which is why your costs are calculated and the observations show under the definition's model observations).(1)

Bedrock reporting region-prefixed inference profile IDs or full ARNs as the model string is expected behavior, and the fix for pricing is exactly what you did: a custom model definition whose regex match_pattern matches the exact string your traces report. User-defined models take priority over the Langfuse-maintained list.(1)

Displaying the name of the matched model definition in the UI and charts instead of the raw ingested model ID is an open feature request, along with the related ask to have Langfuse resolve Bedrock inference profile IDs to their base model IDs or alias multiple IDs to one canonical model so usage charts aren't fragmented.(2) You can follow or add your use case there.

In the meantime, for cleaner analytics you can group cost by model in a dashboard or sort by the model column in the observations table, and save the view for one-click access.(3)

📚 Sources:

1. LLM cost management: how to track, attribute, and control spend - Langfuse
2. Support resolving AWS Bedrock Application Inference Profiles to base model IDs
3. Working with the observations table

---

Have another question? Just tag @inkeep.

[CodeCLS]

Hi @slawekp, did Inkeeps answer help?

[slawekp]

Yeah, it did. At least I know the feature isn't there. Thanks!