diff --git a/litellm-rust/crates/cost/src/completion_response.rs b/litellm-rust/crates/cost/src/completion_response.rs index b812c5542d5..acbdecf2b8e 100644 --- a/litellm-rust/crates/cost/src/completion_response.rs +++ b/litellm-rust/crates/cost/src/completion_response.rs @@ -30,6 +30,7 @@ use crate::realtime_cost::{ partition_results_by_service_tier, }; use crate::responses_usage::ChatUsage; +use crate::search_cost::search_provider_cost_per_query; use crate::speech_cost::count_characters; use crate::together_cost::together_pricing_model; use crate::tool_call_cost_tracking::{DefaultToolRates, ResponseKind as ToolResponseKind}; @@ -811,6 +812,25 @@ pub fn completion_cost_from_response( request.response_time_ms, ), }; + if let CostCall::Search { + number_of_queries, + optional_params, + } = call + { + let search_model = match provider { + Some(provider) if !provider.is_empty() && !model.contains('/') => { + format!("{provider}/search") + } + _ => model.to_owned(), + }; + return search_provider_cost_per_query( + catalog, + &search_model, + provider, + number_of_queries.unwrap_or(1), + optional_params, + ); + } cost_per_token_for_call(catalog, cost_request, call) }, )?; diff --git a/litellm-rust/crates/cost/tests/python_completion_response.rs b/litellm-rust/crates/cost/tests/python_completion_response.rs index c257369d6f9..dab9591ea19 100644 --- a/litellm-rust/crates/cost/tests/python_completion_response.rs +++ b/litellm-rust/crates/cost/tests/python_completion_response.rs @@ -935,6 +935,45 @@ fn search_response_uses_query_list_and_skips_tool_and_additional_charges() { assert!((result.cost.total - 0.0033).abs() < 1e-12); } +#[rstest] +#[case::empty_query_list_bills_nothing("exa_ai", "search", json!({"query": []}), 0.0)] +#[case::single_query_string("exa_ai", "search", json!({"query": "a"}), 0.5)] +#[case::bare_model_reads_the_provider_search_row("tavily", "tavily-search", json!({"query": ["a", "b"]}), 1.0)] +#[case::prefixed_model_keeps_its_own_row("exa_ai", "exa_ai/search", json!({}), 0.5)] +fn search_response_counts_queries_and_resolves_the_search_row_like_python( + #[case] provider: &str, + #[case] model: &str, + #[case] optional_params: Value, + #[case] expected: f64, +) { + let catalog = ModelInfoCatalog::new(HashMap::from([ + ( + "exa_ai/search".to_owned(), + json!({"input_cost_per_query": 0.5}), + ), + ( + "tavily/search".to_owned(), + json!({"input_cost_per_query": 0.5}), + ), + ])); + let response = json!({"model": model}); + let empty = json!({}); + let base = request(Some(&response), Some(model), Some(provider), &empty, &empty); + let result = completion_cost_from_response( + &catalog, + CompletionResponseCostRequest { + input: CompletionInputRequest { + call_type: Some("search"), + optional_params: Some(&optional_params), + ..base.input + }, + ..base + }, + ) + .unwrap(); + assert!((result.cost.total - expected).abs() < 1e-12); +} + #[rstest] #[case("vector_store_search", "vertex_ai/search_api", json!({}), 0.25)] #[case("avector_store_search", "vertex_ai/search_api", json!({}), 0.25)]