ruby_llm 1.0.1 → 1.1.0rc1

data/lib/ruby_llm/providers/gemini/capabilities.rb

@@ -12,10 +12,13 @@ module RubyLLM
         # @return [Integer] the context window size in tokens
         def context_window_for(model_id)
           case model_id
-          when /gemini-2\.0-flash/, /gemini-1\.5-flash/ then 1_048_576
+          when /gemini-2\.5-pro-exp-03-25/, /gemini-2\.0-flash/, /gemini-2\.0-flash-lite/, /gemini-1\.5-flash/, /gemini-1\.5-flash-8b/ # rubocop:disable Layout/LineLength
+            1_048_576
           when /gemini-1\.5-pro/ then 2_097_152
+          when /gemini-embedding-exp/ then 8_192
           when /text-embedding-004/, /embedding-001/ then 2_048
           when /aqa/ then 7_168
+          when /imagen-3/ then nil # No token limit for image generation
           else 32_768 # Sensible default for unknown models
           end
         end
@@ -25,9 +28,13 @@ module RubyLLM
         # @return [Integer] the maximum output tokens
         def max_tokens_for(model_id)
           case model_id
-          when /gemini-2\.0-flash/, /gemini-1\.5/ then 8_192
+          when /gemini-2\.5-pro-exp-03-25/ then 64_000
+          when /gemini-2\.0-flash/, /gemini-2\.0-flash-lite/, /gemini-1\.5-flash/, /gemini-1\.5-flash-8b/, /gemini-1\.5-pro/ # rubocop:disable Layout/LineLength
+            8_192
+          when /gemini-embedding-exp/ then nil # Elastic, supports 3072, 1536, or 768
           when /text-embedding-004/, /embedding-001/ then 768 # Output dimension size for embeddings
           when /aqa/ then 1_024
+          when /imagen-3/ then 4 # Output images
           else 4_096 # Sensible default
           end
         end
@@ -39,8 +46,8 @@ module RubyLLM
           base_price = PRICES.dig(pricing_family(model_id), :input) || default_input_price
           return base_price unless long_context_model?(model_id)
 
-          # Double the price for prompts longer than 128k tokens
-          context_length(model_id) > 128_000 ? base_price * 2 : base_price
+          # Apply different pricing for prompts longer than 128k tokens
+          context_window_for(model_id) > 128_000 ? base_price * 2 : base_price
         end
 
         # Returns the output price per million tokens for the given model
@@ -50,8 +57,8 @@ module RubyLLM
           base_price = PRICES.dig(pricing_family(model_id), :output) || default_output_price
           return base_price unless long_context_model?(model_id)
 
-          # Double the price for prompts longer than 128k tokens
-          context_length(model_id) > 128_000 ? base_price * 2 : base_price
+          # Apply different pricing for prompts longer than 128k tokens
+          context_window_for(model_id) > 128_000 ? base_price * 2 : base_price
         end
 
         # Determines if the model supports vision (image/video) inputs
@@ -59,31 +66,28 @@ module RubyLLM
         # @return [Boolean] true if the model supports vision inputs
         def supports_vision?(model_id)
           return false if model_id.match?(/text-embedding|embedding-001|aqa/)
-          return false if model_id.match?(/gemini-1\.0/)
 
-          model_id.match?(/gemini-[12]\.[05]/)
+          model_id.match?(/gemini|flash|pro|imagen/)
         end
 
         # Determines if the model supports function calling
         # @param model_id [String] the model identifier
         # @return [Boolean] true if the model supports function calling
         def supports_functions?(model_id)
-          return false if model_id.match?(/text-embedding|embedding-001|aqa/)
-          return false if model_id.match?(/flash-lite/)
-          return false if model_id.match?(/gemini-1\.0/)
+          return false if model_id.match?(/text-embedding|embedding-001|aqa|flash-lite|imagen|gemini-2\.0-flash-lite/)
 
-          model_id.match?(/gemini-[12]\.[05]-(?:pro|flash)(?!-lite)/)
+          model_id.match?(/gemini|pro|flash/)
         end
 
         # Determines if the model supports JSON mode
         # @param model_id [String] the model identifier
         # @return [Boolean] true if the model supports JSON mode
         def supports_json_mode?(model_id)
-          return false if model_id.match?(/text-embedding|embedding-001|aqa/)
-          return false if model_id.match?(/gemini-1\.0/)
-          return false if model_id.match?(/gemini-2\.0-flash-lite/)
+          if model_id.match?(/text-embedding|embedding-001|aqa|imagen|gemini-2\.0-flash-lite|gemini-2\.5-pro-exp-03-25/)
+            return false
+          end
 
-          model_id.match?(/gemini-\d/)
+          model_id.match?(/gemini|pro|flash/)
         end
 
         # Formats the model ID into a human-readable display name
@@ -105,23 +109,25 @@ module RubyLLM
         # @param model_id [String] the model identifier
         # @return [Boolean] true if the model supports caching
         def supports_caching?(model_id)
-          return false if model_id.match?(/flash-lite|gemini-1\.0/)
+          if model_id.match?(/flash-lite|gemini-2\.5-pro-exp-03-25|aqa|imagen|text-embedding|embedding-001/)
+            return false
+          end
 
-          model_id.match?(/gemini-[12]\.[05]/)
+          model_id.match?(/gemini|pro|flash/)
         end
 
         # Determines if the model supports tuning
         # @param model_id [String] the model identifier
         # @return [Boolean] true if the model supports tuning
         def supports_tuning?(model_id)
-          model_id.match?(/gemini-1\.5-flash/)
+          model_id.match?(/gemini-1\.5-flash|gemini-1\.5-flash-8b/)
         end
 
         # Determines if the model supports audio inputs
         # @param model_id [String] the model identifier
         # @return [Boolean] true if the model supports audio inputs
         def supports_audio?(model_id)
-          model_id.match?(/gemini-[12]\.[05]/)
+          model_id.match?(/gemini|pro|flash/)
         end
 
         # Returns the type of model (chat, embedding, image)
@@ -129,7 +135,7 @@ module RubyLLM
         # @return [String] the model type
         def model_type(model_id)
           case model_id
-          when /text-embedding|embedding/ then 'embedding'
+          when /text-embedding|embedding|gemini-embedding/ then 'embedding'
           when /imagen/ then 'image'
           else 'chat'
           end
@@ -140,15 +146,17 @@ module RubyLLM
         # @return [String] the model family identifier
         def model_family(model_id) # rubocop:disable Metrics/CyclomaticComplexity,Metrics/MethodLength
           case model_id
+          when /gemini-2\.5-pro-exp-03-25/ then 'gemini25_pro_exp'
           when /gemini-2\.0-flash-lite/ then 'gemini20_flash_lite'
           when /gemini-2\.0-flash/ then 'gemini20_flash'
           when /gemini-1\.5-flash-8b/ then 'gemini15_flash_8b'
           when /gemini-1\.5-flash/ then 'gemini15_flash'
           when /gemini-1\.5-pro/ then 'gemini15_pro'
-          when /gemini-1\.0-pro/ then 'gemini10_pro'
+          when /gemini-embedding-exp/ then 'gemini_embedding_exp'
           when /text-embedding-004/ then 'embedding4'
           when /embedding-001/ then 'embedding1'
           when /aqa/ then 'aqa'
+          when /imagen-3/ then 'imagen3'
           else 'other'
           end
         end
@@ -158,13 +166,15 @@ module RubyLLM
         # @return [Symbol] the pricing family identifier
         def pricing_family(model_id) # rubocop:disable Metrics/CyclomaticComplexity,Metrics/MethodLength
           case model_id
+          when /gemini-2\.5-pro-exp-03-25/ then :pro_2_5 # rubocop:disable Naming/VariableNumber
           when /gemini-2\.0-flash-lite/ then :flash_lite_2 # rubocop:disable Naming/VariableNumber
           when /gemini-2\.0-flash/ then :flash_2 # rubocop:disable Naming/VariableNumber
           when /gemini-1\.5-flash-8b/ then :flash_8b
           when /gemini-1\.5-flash/ then :flash
           when /gemini-1\.5-pro/ then :pro
-          when /gemini-1\.0-pro/ then :pro_1_0 # rubocop:disable Naming/VariableNumber
+          when /gemini-embedding-exp/ then :gemini_embedding
           when /text-embedding|embedding/ then :embedding
+          when /imagen/ then :imagen
           when /aqa/ then :aqa
           else :base
           end
@@ -174,7 +184,7 @@ module RubyLLM
         # @param model_id [String] the model identifier
         # @return [Boolean] true if the model supports long context
         def long_context_model?(model_id)
-          model_id.match?(/gemini-1\.5-(?:pro|flash)/)
+          model_id.match?(/gemini-1\.5-(?:pro|flash)|gemini-1\.5-flash-8b/)
         end
 
         # Returns the context length for the model
@@ -191,43 +201,49 @@ module RubyLLM
             output: 0.40,
             audio_input: 0.70,
             cache: 0.025,
-            cache_storage: 1.00
+            cache_storage: 1.00,
+            grounding_search: 35.00 # per 1K requests after 1.5K free
           },
           flash_lite_2: { # Gemini 2.0 Flash Lite # rubocop:disable Naming/VariableNumber
             input: 0.075,
-            output: 0.30,
-            cache: 0.01875,
-            cache_storage: 1.00
+            output: 0.30
           },
           flash: { # Gemini 1.5 Flash
             input: 0.075,
             output: 0.30,
             cache: 0.01875,
             cache_storage: 1.00,
-            grounding_search: 35.00 # per 1K requests
+            grounding_search: 35.00 # per 1K requests up to 5K per day
           },
           flash_8b: { # Gemini 1.5 Flash 8B
             input: 0.0375,
             output: 0.15,
             cache: 0.01,
             cache_storage: 0.25,
-            grounding_search: 35.00 # per 1K requests
+            grounding_search: 35.00 # per 1K requests up to 5K per day
           },
           pro: { # Gemini 1.5 Pro
             input: 1.25,
             output: 5.0,
             cache: 0.3125,
             cache_storage: 4.50,
-            grounding_search: 35.00 # per 1K requests
+            grounding_search: 35.00 # per 1K requests up to 5K per day
           },
-          pro_1_0: { # Gemini 1.0 Pro # rubocop:disable Naming/VariableNumber
-            input: 0.50,
-            output: 1.50
+          pro_2_5: { # Gemini 2.5 Pro Experimental # rubocop:disable Naming/VariableNumber
+            input: 0.12,
+            output: 0.50
+          },
+          gemini_embedding: { # Gemini Embedding Experimental
+            input: 0.002,
+            output: 0.004
           },
           embedding: { # Text Embedding models
             input: 0.00,
             output: 0.00
           },
+          imagen: { # Imagen 3
+            price: 0.03 # per image
+          },
           aqa: { # AQA model
             input: 0.00,
             output: 0.00

data/lib/ruby_llm/providers/gemini/chat.rb

@@ -4,9 +4,13 @@ module RubyLLM
   module Providers
     module Gemini
       # Chat methods for the Gemini API implementation
-      module Chat # rubocop:disable Metrics/ModuleLength
-        # Must be public for Provider to use
+      module Chat
+        def completion_url
+          "models/#{@model}:generateContent"
+        end
+
         def complete(messages, tools:, temperature:, model:, &block) # rubocop:disable Metrics/MethodLength
+          @model = model
           payload = {
             contents: format_messages(messages),
             generationConfig: {
@@ -20,26 +24,15 @@ module RubyLLM
           @tools = tools
 
           if block_given?
-            stream_completion(model, payload, &block)
+            stream_response payload, &block
           else
-            generate_completion(model, payload)
+            sync_response payload
           end
         end
 
         # Format methods can be private
         private
 
-        def generate_completion(model, payload)
-          url = "models/#{model}:generateContent"
-          response = post(url, payload)
-          result = parse_completion_response(response)
-
-          # If this contains a tool call, log it
-          result.tool_calls.values.first if result.tool_call?
-
-          result
-        end
-
         def format_messages(messages)
           messages.map do |msg|
             {

data/lib/ruby_llm/providers/gemini/images.rb

@@ -5,18 +5,13 @@ module RubyLLM
     module Gemini
       # Image generation methods for the Gemini API implementation
       module Images
-        def images_url(model:)
-          "models/#{model}:predict"
+        def images_url
+          "models/#{@model}:predict"
         end
 
-        def paint(prompt, model:, size:) # rubocop:disable Lint/UnusedMethodArgument
-          payload = render_image_payload(prompt)
-
-          response = post(images_url(model:), payload)
-          parse_image_response(response)
-        end
-
-        def render_image_payload(prompt)
+        def render_image_payload(prompt, model:, size:) # rubocop:disable Metrics/MethodLength
+          RubyLLM.logger.debug "Ignoring size #{size}. Gemini does not support image size customization."
+          @model = model
           {
             instances: [
               {

data/lib/ruby_llm/providers/gemini/streaming.rb

@@ -5,93 +5,52 @@ module RubyLLM
     module Gemini
       # Streaming methods for the Gemini API implementation
       module Streaming
-        # Need to make stream_completion public for chat.rb to access
-        def stream_completion(model, payload, &block) # rubocop:disable Metrics/AbcSize,Metrics/MethodLength
-          url = "models/#{model}:streamGenerateContent?alt=sse"
-          accumulator = StreamAccumulator.new
-
-          post(url, payload) do |req|
-            req.options.on_data = stream_handler(accumulator, &block)
-          end
-
-          # If this is a tool call, immediately execute it and include the result
-          message = accumulator.to_message
-          if message.tool_call? && message.content.to_s.empty? && @tools && !@tools.empty?
-            tool_call = message.tool_calls.values.first
-            tool = @tools[tool_call.name.to_sym]
-
-            if tool
-              tool_result = tool.call(tool_call.arguments)
-              # Create a new chunk with the result
-              result_chunk = Chunk.new(
-                role: :assistant,
-                content: "The result is #{tool_result}",
-                model_id: message.model_id,
-                input_tokens: message.input_tokens,
-                output_tokens: message.output_tokens,
-                tool_calls: message.tool_calls
-              )
-
-              # Add to accumulator and call the block
-              accumulator.add(result_chunk)
-              block.call(result_chunk)
-            end
-          end
+        def stream_url
+          "models/#{@model}:streamGenerateContent?alt=sse"
+        end
 
-          accumulator.to_message
+        def build_chunk(data)
+          Chunk.new(
+            role: :assistant,
+            model_id: extract_model_id(data),
+            content: extract_content(data),
+            input_tokens: extract_input_tokens(data),
+            output_tokens: extract_output_tokens(data),
+            tool_calls: extract_tool_calls(data)
+          )
         end
 
         private
 
-        # Handle streaming
-        def stream_handler(accumulator, &block) # rubocop:disable Metrics/AbcSize,Metrics/CyclomaticComplexity,Metrics/MethodLength,Metrics/PerceivedComplexity
-          to_json_stream do |data| # rubocop:disable Metrics/BlockLength
-            next unless data['candidates']&.any?
-
-            candidate = data['candidates'][0]
-            parts = candidate.dig('content', 'parts')
-            model_id = data['modelVersion']
+        def extract_model_id(data)
+          data['modelVersion']
+        end
 
-            # First attempt to extract tool calls
-            tool_calls = nil
+        def extract_content(data)
+          return nil unless data['candidates']&.any?
 
-            # Check if any part contains a functionCall
-            if parts&.any? { |p| p['functionCall'] }
-              function_part = parts.find { |p| p['functionCall'] }
-              function_data = function_part['functionCall']
+          candidate = data['candidates'][0]
+          parts = candidate.dig('content', 'parts')
+          return nil unless parts
 
-              if function_data && function_data['name']
-                # Create a tool call with proper structure - convert args to JSON string
-                id = SecureRandom.uuid
-                tool_calls = {
-                  id => ToolCall.new(
-                    id: id,
-                    name: function_data['name'],
-                    arguments: JSON.generate(function_data['args']) # Convert Hash to JSON string
-                  )
-                }
-              end
-            end
+          text_parts = parts.select { |p| p['text'] }
+          text_parts.map { |p| p['text'] }.join if text_parts.any?
+        end
 
-            # Extract text content (if any)
-            text = nil
-            if parts
-              text_parts = parts.select { |p| p['text'] }
-              text = text_parts.map { |p| p['text'] }.join if text_parts.any?
-            end
+        def extract_input_tokens(data)
+          data.dig('usageMetadata', 'promptTokenCount')
+        end
 
-            chunk = Chunk.new(
-              role: :assistant,
-              content: text,
-              model_id: model_id,
-              input_tokens: data.dig('usageMetadata', 'promptTokenCount'),
-              output_tokens: data.dig('usageMetadata', 'candidatesTokenCount'),
-              tool_calls: tool_calls
-            )
+        def extract_output_tokens(data)
+          data.dig('usageMetadata', 'candidatesTokenCount')
+        end
 
-            accumulator.add(chunk)
-            block.call(chunk)
-          end
+        def parse_streaming_error(data)
+          error_data = JSON.parse(data)
+          [error_data['error']['code'], error_data['error']['message']]
+        rescue JSON::ParserError => e
+          RubyLLM.logger.debug "Failed to parse streaming error: #{e.message}"
+          [500, "Failed to parse error: #{data}"]
         end
       end
     end

data/lib/ruby_llm/providers/gemini/tools.rb

@@ -54,22 +54,22 @@ module RubyLLM
           {
             name: tool.name,
             description: tool.description,
-            parameters: {
-              type: 'OBJECT',
-              properties: format_parameters(tool.parameters),
-              required: tool.parameters.select { |_, p| p.required }.keys.map(&:to_s)
-            }
-          }
+            parameters: tool.parameters.any? ? format_parameters(tool.parameters) : nil
+          }.compact
         end
 
         # Format tool parameters for Gemini API
         def format_parameters(parameters)
-          parameters.transform_values do |param|
-            {
-              type: param_type_for_gemini(param.type),
-              description: param.description
-            }.compact
-          end
+          {
+            type: 'OBJECT',
+            properties: parameters.transform_values do |param|
+              {
+                type: param_type_for_gemini(param.type),
+                description: param.description
+              }.compact
+            end,
+            required: parameters.select { |_, p| p.required }.keys.map(&:to_s)
+          }
         end
 
         # Convert RubyLLM param types to Gemini API types

data/lib/ruby_llm/providers/gemini.rb

@@ -32,6 +32,10 @@ module RubyLLM
       def slug
         'gemini'
       end
+
+      def configuration_requirements
+        %i[gemini_api_key]
+      end
     end
   end
 end