ruby_llm 1.6.0 → 1.6.2

data/lib/ruby_llm/models.rb

@@ -64,16 +64,12 @@ module RubyLLM
 
           model = Model::Info.new(
             id: model_id,
-            name: model_id.gsub('-', ' ').capitalize,
+            name: model_id.tr('-', ' ').capitalize,
             provider: provider_instance.slug,
             capabilities: %w[function_calling streaming],
             modalities: { input: %w[text image], output: %w[text] },
             metadata: { warning: 'Assuming model exists, capabilities may not be accurate' }
           )
-          if RubyLLM.config.log_assume_model_exists
-            RubyLLM.logger.warn "Assuming model '#{model_id}' exists for provider '#{provider}'. " \
-                                'Capabilities may not be accurately reflected.'
-          end
         else
           model = Models.find model_id, provider
           provider_class = Provider.providers[model.provider.to_sym] || raise(Error,

data/lib/ruby_llm/providers/bedrock/streaming/base.rb

@@ -47,7 +47,7 @@ module RubyLLM
           end
 
           def handle_stream(&block)
-            buffer = String.new
+            buffer = +''
             proc do |chunk, _bytes, env|
               if env && env.status != 200
                 handle_failed_response(chunk, buffer, env)

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

@@ -80,7 +80,7 @@ module RubyLLM
             content: extract_content(data),
             tool_calls: tool_calls,
             input_tokens: data.dig('usageMetadata', 'promptTokenCount'),
-            output_tokens: data.dig('usageMetadata', 'candidatesTokenCount'),
+            output_tokens: calculate_output_tokens(data),
             model_id: data['modelVersion'] || response.env.url.path.split('/')[3].split(':')[0],
             raw: response
           )
@@ -133,6 +133,12 @@ module RubyLLM
           parts = candidate.dig('content', 'parts')
           parts&.any? { |p| p['functionCall'] }
         end
+
+        def calculate_output_tokens(data)
+          candidates = data.dig('usageMetadata', 'candidatesTokenCount') || 0
+          thoughts = data.dig('usageMetadata', 'thoughtsTokenCount') || 0
+          candidates + thoughts
+        end
       end
     end
   end

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

@@ -42,7 +42,10 @@ module RubyLLM
         end
 
         def extract_output_tokens(data)
-          data.dig('usageMetadata', 'candidatesTokenCount')
+          candidates = data.dig('usageMetadata', 'candidatesTokenCount') || 0
+          thoughts = data.dig('usageMetadata', 'thoughtsTokenCount') || 0
+          total = candidates + thoughts
+          total.positive? ? total : nil
         end
 
         def parse_streaming_error(data)

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

@@ -198,11 +198,11 @@ module RubyLLM
             .gsub(/(\d{4}) (\d{2}) (\d{2})/, '\1\2\3')
             .gsub(/^(?:Gpt|Chatgpt|Tts|Dall E) /) { |m| special_prefix_format(m.strip) }
             .gsub(/^O([13]) /, 'O\1-')
-            .gsub(/^O[13] Mini/, '\0'.gsub(' ', '-'))
+            .gsub(/^O[13] Mini/, '\0'.tr(' ', '-'))
             .gsub(/\d\.\d /, '\0'.sub(' ', '-'))
             .gsub(/4o (?=Mini|Preview|Turbo|Audio|Realtime|Transcribe|Tts)/, '4o-')
             .gsub(/\bHd\b/, 'HD')
-            .gsub(/(?:Omni|Text) Moderation/, '\0'.gsub(' ', '-'))
+            .gsub(/(?:Omni|Text) Moderation/, '\0'.tr(' ', '-'))
             .gsub('Text Embedding', 'text-embedding-')
         end
 
@@ -216,7 +216,7 @@ module RubyLLM
         end
 
         def self.normalize_temperature(temperature, model_id)
-          if model_id.match?(/^o\d/)
+          if model_id.match?(/^(o\d|gpt-5)/)
             RubyLLM.logger.debug "Model #{model_id} requires temperature=1.0, ignoring provided value"
             1.0
           elsif model_id.match?(/-search/)

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

@@ -21,10 +21,7 @@ module RubyLLM
           # Only include temperature if it's not nil (some models don't accept it)
           payload[:temperature] = temperature unless temperature.nil?
 
-          if tools.any?
-            payload[:tools] = tools.map { |_, tool| tool_for(tool) }
-            payload[:tool_choice] = 'auto'
-          end
+          payload[:tools] = tools.map { |_, tool| tool_for(tool) } if tools.any?
 
           if schema
             # Use strict mode from schema if specified, default to true

data/lib/ruby_llm/stream_accumulator.rb

@@ -8,7 +8,7 @@ module RubyLLM
     attr_reader :content, :model_id, :tool_calls
 
     def initialize
-      @content = String.new
+      @content = +''
       @tool_calls = {}
       @input_tokens = 0
       @output_tokens = 0
@@ -66,7 +66,7 @@ module RubyLLM
       new_tool_calls.each_value do |tool_call|
         if tool_call.id
           tool_call_id = tool_call.id.empty? ? SecureRandom.uuid : tool_call.id
-          tool_call_arguments = tool_call.arguments.empty? ? String.new : tool_call.arguments
+          tool_call_arguments = tool_call.arguments.empty? ? +'' : tool_call.arguments
           @tool_calls[tool_call.id] = ToolCall.new(
             id: tool_call_id,
             name: tool_call.name,

data/lib/ruby_llm/streaming.rb

@@ -43,7 +43,7 @@ module RubyLLM
     private
 
     def to_json_stream(&)
-      buffer = String.new
+      buffer = +''
       parser = EventStreamParser::Parser.new
 
       create_stream_processor(parser, buffer, &)

data/lib/ruby_llm/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module RubyLLM
-  VERSION = '1.6.0'
+  VERSION = '1.6.2'
 end

data/lib/tasks/aliases.rake

@@ -65,7 +65,7 @@ namespace :aliases do # rubocop:disable Metrics/BlockLength
 
       base_name = Regexp.last_match(1)
       # Normalize to Anthropic naming convention
-      anthropic_name = base_name.gsub('.', '-')
+      anthropic_name = base_name.tr('.', '-')
 
       # Skip if we already have an alias for this
       next if aliases[anthropic_name]
@@ -91,7 +91,7 @@ namespace :aliases do # rubocop:disable Metrics/BlockLength
       # OpenRouter uses "google/" prefix and sometimes different naming
       openrouter_variants = [
         "google/#{model}",
-        "google/#{model.gsub('gemini-', 'gemini-').gsub('.', '-')}",
+        "google/#{model.gsub('gemini-', 'gemini-').tr('.', '-')}",
         "google/#{model.gsub('gemini-', 'gemini-')}"
       ]
 

data/lib/tasks/models_docs.rake

@@ -86,7 +86,7 @@ def generate_models_markdown
 end
 
 def generate_provider_sections
-  RubyLLM::Provider.providers.map do |provider, provider_class|
+  RubyLLM::Provider.providers.filter_map do |provider, provider_class|
     models = RubyLLM.models.by_provider(provider)
     next if models.none?
 
@@ -95,7 +95,7 @@ def generate_provider_sections
 
       #{models_table(models)}
     PROVIDER
-  end.compact.join("\n\n")
+  end.join("\n\n")
 end
 
 def generate_capability_sections
@@ -107,7 +107,7 @@ def generate_capability_sections
     'Batch Processing' => RubyLLM.models.select { |m| m.capabilities.include?('batch') }
   }
 
-  capabilities.map do |capability, models|
+  capabilities.filter_map do |capability, models|
     next if models.none?
 
     <<~CAPABILITY
@@ -115,7 +115,7 @@ def generate_capability_sections
 
       #{models_table(models)}
     CAPABILITY
-  end.compact.join("\n\n")
+  end.join("\n\n")
 end
 
 def generate_modality_sections # rubocop:disable Metrics/PerceivedComplexity

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ruby_llm
 version: !ruby/object:Gem::Version
-  version: 1.6.0
+  version: 1.6.2
 platform: ruby
 authors:
 - Carmine Paolino
@@ -121,12 +121,13 @@ dependencies:
     - - "~>"
       - !ruby/object:Gem::Version
         version: '2'
-description: A delightful Ruby way to work with AI. Tired of juggling different SDKs?
-  RubyLLM provides one beautiful, Ruby-like interface for OpenAI, Anthropic, Gemini,
-  Bedrock, OpenRouter, DeepSeek, Ollama, and any OpenAI-compatible API. Chat (with
-  text, images, audio, PDFs), generate images, create embeddings, use tools (function
-  calling), stream responses, and integrate with Rails effortlessly. Minimal dependencies,
-  maximum developer happiness - just clean Ruby code that works.
+description: One beautiful Ruby API for GPT, Claude, Gemini, and more. Easily build
+  chatbots, AI agents, RAG applications, and content generators. Features chat (text,
+  images, audio, PDFs), image generation, embeddings, tools (function calling), structured
+  output, Rails integration, and streaming. Works with OpenAI, Anthropic, Google Gemini,
+  AWS Bedrock, DeepSeek, Mistral, Ollama (local models), OpenRouter, Perplexity, GPUStack,
+  and any OpenAI-compatible API. Minimal dependencies - just Faraday, Zeitwerk, and
+  Marcel.
 email:
 - carmine@paolino.me
 executables: []
@@ -266,5 +267,5 @@ required_rubygems_version: !ruby/object:Gem::Requirement
 requirements: []
 rubygems_version: 3.6.9
 specification_version: 4
-summary: A single delightful Ruby way to work with AI.
+summary: One beautiful Ruby API for GPT, Claude, Gemini, and more.
 test_files: []