From the source
Perplexity now reports token usage and cost for sandbox extraction calls in Agent API responses, giving developers visibility into per-model consumption.- The new fields appear in usage.tool_calls_details.sandbox.token_usage for tokens and in usage.tool_calls_details.sandbox.cost_usd for the extraction cost.
- Sandbox extraction uses the existing openai/gpt-6-luna pricing, including cache creation, cache read, and reasoning-token dimensions.
- This update corrects response-level usage and cost reporting without changing pricing or backfilling historical responses.




