How to Retrieve Span Costs Using the Python SDK or Alternative Methods?

Kalana P. · 2026-03-10T09:43:06.376Z

Hi Team. Does anyone know how to get the cost of the spans with the Python SDK? Or any other way.

How to Retrieve Span Costs Using the Python SDK or Alternative Methods? | Arize AI Community

RunLLM
APP
·
·
You can retrieve the cost of spans using the Phoenix Python SDK by querying spans and accessing their cost attributes. The SDK provides methods to extract spans as a DataFrame, which includes cost information if cost tracking is enabled and the required span attributes are present (e.g., `llm.token_count.prompt`, `llm.token_count.completion`, `llm.token_count.total`, `llm.model_name`, and `llm.provider`)—these are automatically captured with supported auto-instrumentation or can be manually added. Cost data is then available in the DataFrame for further analysis or export. For more details, see the <https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking|cost tracking documentation> and <https://arize.com/docs/phoenix/tracing/how-to-tracing/importing-and-exporting-traces/extract-data-from-spans#running-span-queries|span querying guide>. Example code to pull spans (including cost) as a DataFrame: ``` from phoenix.client import Client client = Client() spans = client.spans.get_spans_dataframe(project_identifier="default") # The resulting DataFrame will include cost columns if cost tracking is configured ``` Would you like more detail on filtering or analyzing cost data? Sources: - <https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking|Cost Tracking - Phoenix> - <https://arize.com/docs/phoenix/tracing/how-to-tracing/importing-and-exporting-traces/extract-data-from-spans#running-span-queries|span querying guide>
💡 Hint: Mention RunLLM in the thread for followups.
Kalana P.
·
·
Thanks will check this
Kalana P.
·
·
the token count I can get. The issue is the actual cost. In the Phoenix dashboard, we can configure each model's costs. So is it possible to get that cost?
Kalana P.
·
·
RunLLM
RunLLM
APP
·
·
Yes, Phoenix automatically calculates and stores the actual cost of each span based on the model pricing configured in the dashboard (Settings → Models). When you extract spans using the Python SDK (e.g., with `get_spans_dataframe()`), the resulting DataFrame includes columns for cost (such as total, prompt, and completion costs) if cost tracking is enabled and model pricing is set up in the UI. You do not need to manually calculate costs from token counts—the SDK will return the computed costs as long as the required attributes and pricing are present. See <https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking#viewing-cost-data|cost tracking documentation> and <https://arize.com/docs/phoenix/tracing/how-to-tracing/importing-and-exporting-traces/extract-data-from-spans|exporting spans>. Would you like a code example or more detail on how to access these cost columns? Sources: - <https://arize.com/docs/phoenix/tracing/how-to-tracing/importing-and-exporting-traces/extract-data-from-spans|exporting spans>
💡 Hint: Mention RunLLM in the thread for followups.
Kalana P.
·
·
RunLLM the cost is not there. I checked available columns in the get_spans_dataframe() I cant see any cost cols
RunLLM
APP
·
·
If cost columns are missing from your `get_spans_dataframe()` output, it means either cost tracking is not enabled, required span attributes are missing, or model pricing is not properly configured in Phoenix. The DataFrame will only include cost columns if spans have attributes like `llm.token_count.prompt`, `llm.token_count.completion`, `llm.token_count.total`, `llm.model_name`, and `llm.provider`, and if the model pricing table in Settings → Models is set up for those models. See <https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking#model-pricing-configuration|cost tracking requirements>. If these are set up and you still don't see cost columns, your Phoenix version may not support this feature or there may be a mismatch in model/provider names. Double-check your span attributes and model pricing configuration. Would you like step-by-step troubleshooting? Sources: - <https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking#model-pricing-configuration|cost tracking requirements>
💡 Hint: Mention RunLLM in the thread for followups.
Kalana P.
·
·
RunLLM in the UI, I can see the cost and stuff. but in the Python SDK, I cant see that specific column. what will the column name be?
RunLLM
APP
·
·
In the Python SDK, the cost columns are typically named `cost_total`, `cost_prompt`, and `cost_completion` in the DataFrame returned by `get_spans_dataframe()`, provided cost tracking is enabled and configured. If these columns are not present, it means the cost attributes were not computed or included for those spans, even if the UI displays them (the UI may aggregate or compute costs differently). Would you like more detail on how to ensure these columns appear or how to map UI fields to DataFrame columns? Sources: - <https://github.com/Arize-ai/openinference/blob/main/python/openinference-semantic-conventions/src/openinference/semconv/trace/__init__.py|https://github.com/Arize-ai/openinference/blob/main/python/openinference-semantic-conventions/src/openinference/semconv/trace/__init__.py> - <https://github.com/Arize-ai/openinference/blob/main/js/packages/openinference-semantic-conventions/src/trace/SemanticConventions.ts|https://github.com/Arize-ai/openinference/blob/main/js/packages/openinference-semantic-conventions/src/trace/SemanticConventions.ts>
💡 Hint: Mention RunLLM in the thread for followups.
Kalana P.
·
·
how to configure that in to python SDK. RunLLM
RunLLM
APP
·
·
You do not configure cost tracking directly in the Python SDK; instead, you must ensure that your spans include the required attributes (`llm.token_count.prompt`, `llm.token_count.completion`, `llm.token_count.total`, `llm.model_name`, and `llm.provider`) and that model pricing is set up in the Phoenix UI under Settings → Models. The SDK will then automatically include cost columns in the DataFrame if these prerequisites are met. See <https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking#model-pricing-configuration|cost tracking setup>. Would you like step-by-step instructions for setting up model pricing or ensuring your spans have the correct attributes? Sources: - <https://arize.com/docs/phoenix/tracing/how-to-tracing/cost-tracking#model-pricing-configuration|cost tracking setup>
💡 Hint: Mention RunLLM in the thread for followups.
Kalana P.
·
·
I can see the cost per span in the UI, Phoenix. issue is with the SDK. I cant see that. RunLLM
RunLLM
APP
·
·
If you see cost per span in the Phoenix UI but not in the Python SDK DataFrame, it's likely because the SDK's `get_spans_dataframe()` does not currently expose the computed cost columns, even though the backend and UI display them. This is a known limitation—cost is calculated and stored in Phoenix, but the SDK may not yet surface these fields in its DataFrame output. See <https://arize.com/docs/phoenix/tracing/how-to-tracing/importing-and-exporting-traces/extract-data-from-spans|exporting spans documentation> and related <https://github.com/Arize-ai/phoenix/issues/6271|GitHub discussions>. For now, you may need to use the Phoenix UI or GraphQL API to access cost details per span, as the SDK does not guarantee these columns will be present in all versions. Would you like more detail on using the GraphQL API or other workarounds? Sources: - <https://arize.com/docs/phoenix/tracing/how-to-tracing/importing-and-exporting-traces/extract-data-from-spans|exporting spans documentation> - <https://github.com/Arize-ai/phoenix/issues/6271|GitHub discussions>
💡 Hint: Mention RunLLM in the thread for followups.
Kalana P.
·
·
see is it possible to use GraphQL
Kalana P.
·
·
RunLLM

17 comments