llm_cost_tracker 0.7.2 → 0.8.0

data/lib/llm_cost_tracker/parsers/openai_usage.rb

@@ -1,8 +1,14 @@
 # frozen_string_literal: true
 
+require_relative "openai_service_charges"
+
 module LlmCostTracker
   module Parsers
     module OpenaiUsage
+      include OpenaiServiceCharges
+
+      OPENAI_DATA_RESIDENCY_HOST_PATTERN = /\A[a-z]{2,3}\.api\.openai\.com\z/
+
       private
 
       def parse_openai_usage(request_url:, request_body:, response_status:, response_body:)
@@ -27,7 +33,8 @@ module LlmCostTracker
           ),
           model: model,
           token_usage: token_usage(usage: usage, cache_read: cache_read),
-          usage_source: :response
+          usage_source: :response,
+          service_line_items: openai_service_line_items(response)
         )
       end
 
@@ -35,8 +42,7 @@ module LlmCostTracker
         return nil unless response_status == 200
 
         request = safe_json_parse(request_body)
-        model =
-          find_event_value(events) { |data| data["model"] || data.dig("response", "model") } || request["model"]
+        model = find_event_value(events) { |data| data["model"] || data.dig("response", "model") } || request["model"]
         usage = detect_stream_usage(events)
         response_id = find_event_value(events) { |data| data["id"] || data.dig("response", "id") }
         pricing_mode = pricing_mode(
@@ -44,6 +50,7 @@ module LlmCostTracker
           model: model,
           service_tier: stream_pricing_mode(events) || request["service_tier"]
         )
+        service_line_items = openai_stream_service_line_items(events)
 
         if usage
           cache_read = cache_read_input_tokens(usage)
@@ -54,14 +61,16 @@ module LlmCostTracker
             model: model,
             token_usage: token_usage(usage: usage, cache_read: cache_read),
             stream: true,
-            usage_source: :stream_final
+            usage_source: :stream_final,
+            service_line_items: service_line_items
           )
         else
           build_unknown_stream_usage(
             provider: provider_for(request_url),
             model: model,
             provider_response_id: response_id,
-            pricing_mode: pricing_mode
+            pricing_mode: pricing_mode,
+            service_line_items: service_line_items
           )
         end
       end
@@ -88,7 +97,7 @@ module LlmCostTracker
 
       def openai_regional_processing?(request_url:, model:)
         uri = parsed_uri(request_url)
-        return false unless %w[us.api.openai.com eu.api.openai.com].include?(uri&.host.to_s.downcase)
+        return false unless uri&.host.to_s.downcase.match?(OPENAI_DATA_RESIDENCY_HOST_PATTERN)
 
         openai_data_residency_model?(model)
       end
@@ -98,36 +107,54 @@ module LlmCostTracker
       end
 
       def token_usage(usage:, cache_read:)
+        audio_input = audio_input_tokens(usage)
+        audio_output = audio_output_tokens(usage)
+
         TokenUsage.build(
-          input_tokens: regular_input_tokens(usage: usage, cache_read: cache_read),
-          output_tokens: (usage["completion_tokens"] || usage["output_tokens"]).to_i,
-          total_tokens: total_tokens(usage: usage, cache_read: cache_read),
+          input_tokens: regular_input_tokens(usage: usage, cache_read: cache_read, audio_input: audio_input),
+          output_tokens: regular_output_tokens(usage: usage, audio_output: audio_output),
+          total_tokens: usage["total_tokens"],
           cache_read_input_tokens: cache_read,
+          audio_input_tokens: audio_input,
+          audio_output_tokens: audio_output,
           hidden_output_tokens: hidden_output_tokens(usage)
         )
       end
 
-      def regular_input_tokens(usage:, cache_read:)
-        [(usage["prompt_tokens"] || usage["input_tokens"]).to_i - cache_read.to_i, 0].max
+      def regular_input_tokens(usage:, cache_read:, audio_input:)
+        [(usage["prompt_tokens"] || usage["input_tokens"]).to_i - cache_read - audio_input, 0].max
+      end
+
+      def regular_output_tokens(usage:, audio_output:)
+        [(usage["completion_tokens"] || usage["output_tokens"]).to_i - audio_output, 0].max
       end
 
       def cache_read_input_tokens(usage)
-        details = usage["prompt_tokens_details"] || usage["input_tokens_details"] || {}
-        details["cached_tokens"]
+        details = input_token_details(usage)
+        details["cached_tokens"].to_i
+      end
+
+      def audio_input_tokens(usage)
+        details = input_token_details(usage)
+        details["audio_tokens"].to_i
       end
 
       def hidden_output_tokens(usage)
-        details = usage["completion_tokens_details"] || usage["output_tokens_details"] || {}
-        details["reasoning_tokens"]
+        details = output_token_details(usage)
+        details["reasoning_tokens"].to_i
       end
 
-      def total_tokens(usage:, cache_read:)
-        total = usage["total_tokens"]
-        return total.to_i unless total.nil?
+      def audio_output_tokens(usage)
+        details = output_token_details(usage)
+        details["audio_tokens"].to_i
+      end
+
+      def input_token_details(usage)
+        usage["prompt_tokens_details"] || usage["input_tokens_details"] || usage["input_token_details"] || {}
+      end
 
-        regular_input_tokens(usage: usage, cache_read: cache_read) +
-          cache_read.to_i +
-          (usage["completion_tokens"] || usage["output_tokens"]).to_i
+      def output_token_details(usage)
+        usage["completion_tokens_details"] || usage["output_tokens_details"] || usage["output_token_details"] || {}
       end
     end
   end

data/lib/llm_cost_tracker/parsers/sse.rb

@@ -12,7 +12,7 @@ module LlmCostTracker
         def parse(body)
           return [] if body.blank?
 
-          return parse_json_array(body) if body.lstrip.start_with?("[")
+          return parse_json_array(body) if body.match?(/\A\s*\[/)
 
           parse_event_stream(body)
         end

data/lib/llm_cost_tracker/parsers.rb

@@ -13,7 +13,7 @@ module LlmCostTracker
     def find_for_provider(provider)
       provider_name = provider.to_s.downcase
       BUILT_INS.find do |parser|
-        Array(parser.provider_names).map { |name| name.to_s.downcase }.include?(provider_name)
+        parser.provider_names.include?(provider_name)
       end
     end
   end

data/lib/llm_cost_tracker/prices.json

@@ -1,6 +1,6 @@
 {
   "metadata": {
-    "updated_at": "2026-05-01",
+    "updated_at": "2026-05-02",
     "currency": "USD",
     "unit": "1M tokens",
     "source_urls": [
@@ -15,13 +15,23 @@
     "schema_version": 1,
     "min_gem_version": "0.4.0"
   },
+  "service_charges": {
+    "anthropic": {
+      "web_search_request": 10.0,
+      "code_execution_hour": 0.05
+    },
+    "openai": {
+      "web_search_request": 10.0,
+      "file_search_call": 2.5
+    }
+  },
   "models": {
     "anthropic/claude-haiku-4-5": {
       "input": 1.0,
       "output": 5.0,
       "cache_read_input": 0.1,
       "cache_write_input": 1.25,
-      "cache_write_1h_input": 2.0,
+      "cache_write_extended_input": 2.0,
       "batch_input": 0.5,
       "batch_output": 2.5
     },
@@ -30,7 +40,7 @@
       "output": 75.0,
       "cache_read_input": 1.5,
       "cache_write_input": 18.75,
-      "cache_write_1h_input": 30.0,
+      "cache_write_extended_input": 30.0,
       "batch_input": 7.5,
       "batch_output": 37.5
     },
@@ -39,7 +49,7 @@
       "output": 75.0,
       "cache_read_input": 1.5,
       "cache_write_input": 18.75,
-      "cache_write_1h_input": 30.0,
+      "cache_write_extended_input": 30.0,
       "batch_input": 7.5,
       "batch_output": 37.5
     },
@@ -48,7 +58,7 @@
       "output": 25.0,
       "cache_read_input": 0.5,
       "cache_write_input": 6.25,
-      "cache_write_1h_input": 10.0,
+      "cache_write_extended_input": 10.0,
       "batch_input": 2.5,
       "batch_output": 12.5
     },
@@ -57,24 +67,24 @@
       "output": 25.0,
       "cache_read_input": 0.5,
       "cache_write_input": 6.25,
-      "cache_write_1h_input": 10.0,
+      "cache_write_extended_input": 10.0,
       "batch_input": 2.5,
       "batch_output": 12.5,
       "data_residency_input": 5.5,
       "data_residency_cache_write_input": 6.875,
-      "data_residency_cache_write_1h_input": 11.0,
+      "data_residency_cache_write_extended_input": 11.0,
       "data_residency_cache_read_input": 0.55,
       "data_residency_output": 27.5,
       "data_residency_batch_input": 2.75,
       "data_residency_batch_output": 13.75,
       "fast_input": 30.0,
       "fast_cache_write_input": 37.5,
-      "fast_cache_write_1h_input": 60.0,
+      "fast_cache_write_extended_input": 60.0,
       "fast_cache_read_input": 3.0,
       "fast_output": 150.0,
       "fast_data_residency_input": 33.0,
       "fast_data_residency_cache_write_input": 41.25,
-      "fast_data_residency_cache_write_1h_input": 66.0,
+      "fast_data_residency_cache_write_extended_input": 66.0,
       "fast_data_residency_cache_read_input": 3.3,
       "fast_data_residency_output": 165.0
     },
@@ -83,12 +93,12 @@
       "output": 25.0,
       "cache_read_input": 0.5,
       "cache_write_input": 6.25,
-      "cache_write_1h_input": 10.0,
+      "cache_write_extended_input": 10.0,
       "batch_input": 2.5,
       "batch_output": 12.5,
       "data_residency_input": 5.5,
       "data_residency_cache_write_input": 6.875,
-      "data_residency_cache_write_1h_input": 11.0,
+      "data_residency_cache_write_extended_input": 11.0,
       "data_residency_cache_read_input": 0.55,
       "data_residency_output": 27.5,
       "data_residency_batch_input": 2.75,
@@ -99,7 +109,7 @@
       "output": 15.0,
       "cache_read_input": 0.3,
       "cache_write_input": 3.75,
-      "cache_write_1h_input": 6.0,
+      "cache_write_extended_input": 6.0,
       "batch_input": 1.5,
       "batch_output": 7.5
     },
@@ -108,7 +118,7 @@
       "output": 15.0,
       "cache_read_input": 0.3,
       "cache_write_input": 3.75,
-      "cache_write_1h_input": 6.0,
+      "cache_write_extended_input": 6.0,
       "batch_input": 1.5,
       "batch_output": 7.5
     },
@@ -117,12 +127,12 @@
       "output": 15.0,
       "cache_read_input": 0.3,
       "cache_write_input": 3.75,
-      "cache_write_1h_input": 6.0,
+      "cache_write_extended_input": 6.0,
       "batch_input": 1.5,
       "batch_output": 7.5,
       "data_residency_input": 3.3,
       "data_residency_cache_write_input": 4.125,
-      "data_residency_cache_write_1h_input": 6.6,
+      "data_residency_cache_write_extended_input": 6.6,
       "data_residency_cache_read_input": 0.33,
       "data_residency_output": 16.5,
       "data_residency_batch_input": 1.65,
@@ -134,7 +144,9 @@
       "output": 0.4,
       "batch_input": 0.05,
       "batch_output": 0.2,
-      "batch_cache_read_input": 0.025
+      "batch_cache_read_input": 0.025,
+      "audio_input": 0.7,
+      "batch_audio_input": 0.35
     },
     "gemini/gemini-2.0-flash-lite": {
       "input": 0.075,
@@ -154,7 +166,11 @@
       "flex_cache_read_input": 0.03,
       "priority_input": 0.54,
       "priority_output": 4.5,
-      "priority_cache_read_input": 0.054
+      "priority_cache_read_input": 0.054,
+      "audio_input": 1.0,
+      "batch_audio_input": 0.5,
+      "flex_audio_input": 0.5,
+      "priority_audio_input": 1.8
     },
     "gemini/gemini-2.5-flash-lite": {
       "input": 0.1,
@@ -168,7 +184,11 @@
       "flex_cache_read_input": 0.01,
       "priority_input": 0.18,
       "priority_output": 0.72,
-      "priority_cache_read_input": 0.018
+      "priority_cache_read_input": 0.018,
+      "audio_input": 0.3,
+      "batch_audio_input": 0.15,
+      "flex_audio_input": 0.15,
+      "priority_audio_input": 0.54
     },
     "gemini/gemini-2.5-pro": {
       "input": 1.25,
@@ -309,6 +329,71 @@
       "priority_output": 1.0,
       "priority_cache_read_input": 0.125
     },
+    "openai/gpt-4o-realtime-preview": {
+      "input": 5.0,
+      "cache_read_input": 2.5,
+      "audio_input": 40.0,
+      "output": 20.0,
+      "audio_output": 80.0
+    },
+    "openai/gpt-4o-mini-realtime-preview": {
+      "input": 0.6,
+      "cache_read_input": 0.3,
+      "audio_input": 10.0,
+      "output": 2.4,
+      "audio_output": 20.0
+    },
+    "openai/gpt-realtime": {
+      "input": 4.0,
+      "cache_read_input": 0.4,
+      "audio_input": 32.0,
+      "output": 16.0,
+      "audio_output": 64.0
+    },
+    "openai/gpt-realtime-1.5": {
+      "input": 4.0,
+      "cache_read_input": 0.4,
+      "audio_input": 32.0,
+      "output": 16.0,
+      "audio_output": 64.0
+    },
+    "openai/gpt-realtime-mini": {
+      "input": 0.6,
+      "cache_read_input": 0.06,
+      "audio_input": 10.0,
+      "output": 2.4,
+      "audio_output": 20.0
+    },
+    "openai/gpt-audio-1.5": {
+      "input": 2.5,
+      "audio_input": 32.0,
+      "output": 10.0,
+      "audio_output": 64.0
+    },
+    "openai/gpt-audio-mini": {
+      "input": 0.6,
+      "audio_input": 10.0,
+      "output": 2.4,
+      "audio_output": 20.0
+    },
+    "openai/gpt-audio": {
+      "input": 2.5,
+      "audio_input": 32.0,
+      "output": 10.0,
+      "audio_output": 64.0
+    },
+    "openai/gpt-4o-audio-preview": {
+      "input": 2.5,
+      "audio_input": 40.0,
+      "output": 10.0,
+      "audio_output": 80.0
+    },
+    "openai/gpt-4o-mini-audio-preview": {
+      "input": 0.15,
+      "audio_input": 10.0,
+      "output": 0.6,
+      "audio_output": 20.0
+    },
     "openai/gpt-5": {
       "input": 1.25,
       "output": 10.0,
@@ -672,7 +757,7 @@
     "anthropic/claude-haiku-3-5": {
       "input": 0.8,
       "cache_write_input": 1.0,
-      "cache_write_1h_input": 1.6,
+      "cache_write_extended_input": 1.6,
       "cache_read_input": 0.08,
       "output": 4.0,
       "batch_input": 0.4,
@@ -681,7 +766,7 @@
     "anthropic/claude-haiku-3": {
       "input": 0.25,
       "cache_write_input": 0.3,
-      "cache_write_1h_input": 0.5,
+      "cache_write_extended_input": 0.5,
       "cache_read_input": 0.03,
       "output": 1.25,
       "batch_input": 0.125,

data/lib/llm_cost_tracker/pricing/effective_prices.rb

@@ -1,18 +1,17 @@
 # frozen_string_literal: true
 
-require_relative "components"
+require_relative "../billing/components"
 
 module LlmCostTracker
   module Pricing
     module EffectivePrices
       class << self
         def call(usage:, prices:, pricing_mode:)
-          quantities = usage.price_quantities
           context_tier = context_tier?(usage: usage, prices: prices)
 
-          Pricing::COMPONENTS.to_h do |component|
-            price_key = component.price_key
-            tokens = quantities.fetch(price_key)
+          Billing::Components::TOKEN_PRICED.to_h do |component|
+            price_key = component.key
+            tokens = usage.public_send(component.token_key)
             price = if tokens.positive?
                       price_for(
                         prices: prices,
@@ -30,29 +29,67 @@ module LlmCostTracker
         private
 
         def price_for(prices:, key:, pricing_mode:, context_tier:)
-          mode = Pricing.normalize_mode(pricing_mode)
-          return contextual_price(prices: prices, key: key, context_tier: context_tier) unless mode
+          return contextual_price(prices: prices, key: key, context_tier: context_tier) unless pricing_mode
 
-          contextual_price(prices: prices, key: :"#{mode}_#{key}", context_tier: context_tier) ||
-            derived_mode_price(prices: prices, key: key, mode: mode, context_tier: context_tier)
+          orderings = mode_orderings_for(pricing_mode)
+          orderings.each do |mode|
+            direct = contextual_price(prices: prices, key: :"#{mode}_#{key}", context_tier: context_tier)
+            return direct if direct
+          end
+          return nil if %i[input output].include?(key)
+
+          derived_mode_price(prices: prices, key: key, modes: orderings, context_tier: context_tier)
         end
 
+        def mode_orderings_for(pricing_mode)
+          mode_string = pricing_mode.to_s
+          return [mode_string] unless mode_string.include?("_")
+
+          tokens = tokenize_mode(mode_string)
+          return [mode_string] if tokens.size <= 1
+
+          [mode_string, *tokens.permutation.map { |permutation| permutation.join("_") }].uniq
+        end
+
+        def tokenize_mode(mode_string)
+          remaining = mode_string.dup
+          tokens = []
+          loop do
+            break if remaining.empty?
+
+            compound = COMPOUND_MODE_TOKENS.find { |token| remaining == token || remaining.start_with?("#{token}_") }
+            if compound
+              tokens << compound
+              remaining = remaining.delete_prefix(compound).delete_prefix("_")
+            else
+              first, _, rest = remaining.partition("_")
+              tokens << first
+              remaining = rest
+            end
+          end
+          tokens
+        end
+
+        COMPOUND_MODE_TOKENS = %w[data_residency].freeze
+        private_constant :COMPOUND_MODE_TOKENS
+
         def contextual_price(prices:, key:, context_tier:)
           return prices[key] unless context_tier
 
           prices[:"above_context_#{key}"]
         end
 
-        def derived_mode_price(prices:, key:, mode:, context_tier:)
+        def derived_mode_price(prices:, key:, modes:, context_tier:)
           standard_price = contextual_price(prices: prices, key: key, context_tier: context_tier)
-          return nil unless standard_price
-
-          base_key = key == :output ? :output : :input
-          base_price = contextual_price(prices: prices, key: base_key, context_tier: context_tier)
-          mode_base_price = contextual_price(prices: prices, key: :"#{mode}_#{base_key}", context_tier: context_tier)
-          return nil unless base_price && mode_base_price
+          base_price = contextual_price(prices: prices, key: :input, context_tier: context_tier)
+          return nil unless standard_price && base_price
+          return nil if base_price.zero?
 
-          standard_price * (mode_base_price.to_f / base_price)
+          modes.each do |mode|
+            mode_base_price = contextual_price(prices: prices, key: :"#{mode}_input", context_tier: context_tier)
+            return standard_price * (mode_base_price / base_price) if mode_base_price
+          end
+          nil
         end
 
         def context_tier?(usage:, prices:)
@@ -62,8 +99,9 @@ module LlmCostTracker
           input_tokens = usage.input_tokens +
                          usage.cache_read_input_tokens +
                          usage.cache_write_input_tokens +
-                         usage.cache_write_1h_input_tokens
-          input_tokens > threshold.to_i
+                         usage.cache_write_extended_input_tokens +
+                         usage.audio_input_tokens
+          input_tokens > threshold
         end
       end
     end

data/lib/llm_cost_tracker/pricing/explainer.rb

@@ -1,5 +1,6 @@
 # frozen_string_literal: true
 
+require_relative "../token_usage"
 require_relative "effective_prices"
 
 module LlmCostTracker
@@ -33,7 +34,7 @@ module LlmCostTracker
 
     module Explainer
       class << self
-        def call(provider:, model:, token_usage:, pricing_mode: nil)
+        def call(provider:, model:, tokens:, pricing_mode: nil)
           match = Lookup.call(provider: provider, model: model)
 
           explanation(
@@ -41,7 +42,7 @@ module LlmCostTracker
             model: model,
             pricing_mode: pricing_mode,
             match: match,
-            usage: token_usage
+            usage: TokenUsage.build_from_tokens(tokens)
           )
         end
 
@@ -50,9 +51,7 @@ module LlmCostTracker
         def explanation(provider:, model:, pricing_mode:, match:, usage:)
           prices = match&.prices
           pricing_mode = Pricing.normalize_mode(pricing_mode)
-          effective = if prices && usage
-                        EffectivePrices.call(usage: usage, prices: prices, pricing_mode: pricing_mode)
-                      end
+          effective = EffectivePrices.call(usage: usage, prices: prices, pricing_mode: pricing_mode) if prices
 
           Explanation.new(
             provider: provider.to_s,