Accurate sandbox extraction usage and costs
By Steven Van ·
Successful sandbox extraction calls now count towards Agent API response totals, with pricing unchanged and no historical backfill.
Successful pplx_sdk.llm.extract calls inside the sandbox tool now have their model token usage and cost included in Perplexity Agent API responses.
Where usage and costs appear
Per-model token counts appear in usage.tool_calls_details.sandbox.token_usage. Extraction costs appear in both usage.tool_calls_details.sandbox.cost_usd and usage.cost.tool_calls_cost_details.sandbox, and contribute to usage.cost.total_cost.
Existing pricing still applies
The update uses existing openai/gpt-6-luna pricing, including cache creation, cache read, and reasoning-token dimensions. It corrects response-level reporting without changing pricing or backfilling historical responses.