llm_cost_tracker 0.15.0 → 0.15.2

data/lib/llm_cost_tracker/providers/openai/parser.rb

@@ -8,7 +8,7 @@ module LlmCostTracker
         include StreamParser
 
         TRACKED_ENDPOINTS = %w[
-          chat/completions completions embeddings moderations responses responses/compact
+          chat/completions completions decisions embeddings moderations responses responses/compact
           audio/transcriptions audio/translations audio/speech
           images/generations images/edits images/variations
         ].freeze

data/lib/llm_cost_tracker/providers/openai/response_parser.rb

@@ -14,6 +14,7 @@ module LlmCostTracker
         PENDING_STATUSES = %w[queued in_progress].freeze
         RETRIEVE_PATH_PATTERN = %r{/(?:responses|agent)/resp_}
         TRANSCRIPTION_PATH_PATTERN = %r{/audio/(?:transcriptions|translations)\z}
+        UNBILLED_DECISION_USAGE = %i[output_tokens output_tokens_details total_tokens].freeze
 
         class << self
           def included(base) = base.include(StreamParser)
@@ -29,7 +30,7 @@ module LlmCostTracker
           end
 
           def event_from_response(response:, request:, provider:, host:, usage_source:, pricing_mode: nil)
-            usage = response["usage"]&.deep_symbolize_keys
+            usage = billed_usage(response, provider)
             return nil if usage.nil? || pending?(response)
 
             model = response["model"] || request["model"]
@@ -73,6 +74,15 @@ module LlmCostTracker
           def pending?(response)
             PENDING_STATUSES.include?(response["status"].to_s)
           end
+
+          def billed_usage(response, provider)
+            usage = response["usage"]&.deep_symbolize_keys
+            return usage unless usage && decision?(response, provider)
+
+            usage.except(*UNBILLED_DECISION_USAGE)
+          end
+
+          def decision?(response, provider) = provider.to_s == "openai" && response.key?("answers")
         end
 
         def parse(request_url:, request_body:, response_status:, response_body:, **)

data/lib/llm_cost_tracker/providers/openai/usage_extractor.rb

@@ -16,12 +16,14 @@ module LlmCostTracker
           def token_usage(usage, model: nil, default_to_image: ModelFamilies.image_output?(model))
             cache_read = cache_read_input_tokens(usage)
             cache_write = cache_write_input_tokens(usage)
+            extended = cache_write_extended_input_tokens(usage).clamp(0, [cache_write, 0].max)
             Usage::TokenUsage.build(
               **input_counts(usage, cached: cache_read + cache_write),
               **output_counts(usage, default_to_image: default_to_image),
               total_tokens: usage[:total_tokens],
               cache_read_input_tokens: cache_read,
-              cache_write_input_tokens: cache_write,
+              cache_write_input_tokens: cache_write - extended,
+              cache_write_extended_input_tokens: extended,
               hidden_output_tokens: hidden_output_tokens(usage)
             )
           end
@@ -48,6 +50,12 @@ module LlmCostTracker
 
           def cache_read_input_tokens(usage) = detail(usage, INPUT_DETAIL_KEYS, :cached_tokens)
           def cache_write_input_tokens(usage) = detail(usage, INPUT_DETAIL_KEYS, :cache_write_tokens)
+
+          def cache_write_extended_input_tokens(usage)
+            details = INPUT_DETAIL_KEYS.filter_map { |key| usage.dig(key, :cache_write_tokens_details) }.first
+            details.is_a?(Hash) ? details[:ttl_12h_tokens].to_i : 0
+          end
+
           def hidden_output_tokens(usage)    = detail(usage, OUTPUT_DETAIL_KEYS, :reasoning_tokens)
           def audio_input_tokens(usage)      = detail(usage, INPUT_DETAIL_KEYS, :audio_tokens)
           def audio_output_tokens(usage)     = detail(usage, OUTPUT_DETAIL_KEYS, :audio_tokens)

data/lib/llm_cost_tracker/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module LlmCostTracker
-  VERSION = "0.15.0"
+  VERSION = "0.15.2"
 end

data/llm_cost_tracker.gemspec

@@ -49,7 +49,7 @@ Gem::Specification.new do |spec|
   spec.add_dependency "faraday", ">= 2.0", "< 3.0"
   spec.add_dependency "railties", ">= 8.0", "< 9.0"
 
-  spec.add_development_dependency "anthropic", "~> 1.61"
+  spec.add_development_dependency "anthropic", ">= 1.61", "< 3"
   spec.add_development_dependency "aws-eventstream", "~> 1.4"
   spec.add_development_dependency "nokogiri", "~> 1.19"
   spec.add_development_dependency "openai", "~> 0.80"

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: llm_cost_tracker
 version: !ruby/object:Gem::Version
-  version: 0.15.0
+  version: 0.15.2
 platform: ruby
 authors:
 - Sergii Khomenko
@@ -107,16 +107,22 @@ dependencies:
   name: anthropic
   requirement: !ruby/object:Gem::Requirement
     requirements:
-    - - "~>"
+    - - ">="
       - !ruby/object:Gem::Version
         version: '1.61'
+    - - "<"
+      - !ruby/object:Gem::Version
+        version: '3'
   type: :development
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
-    - - "~>"
+    - - ">="
       - !ruby/object:Gem::Version
         version: '1.61'
+    - - "<"
+      - !ruby/object:Gem::Version
+        version: '3'
 - !ruby/object:Gem::Dependency
   name: aws-eventstream
   requirement: !ruby/object:Gem::Requirement