Base mistral implementation

Author: OriPekelman

lib/ruby_llm.rb

@@ -17,7 +17,8 @@
   'api' => 'API',
   'deepseek' => 'DeepSeek',
   'bedrock' => 'Bedrock',
-  'openrouter' => 'OpenRouter'
+  'openrouter' => 'OpenRouter',
+  'mistral' => 'Mistral'
 )
 loader.ignore("#{__dir__}/tasks")
 loader.ignore("#{__dir__}/ruby_llm/railtie")
@@ -82,6 +83,7 @@ def logger
 RubyLLM::Provider.register :bedrock, RubyLLM::Providers::Bedrock
 RubyLLM::Provider.register :openrouter, RubyLLM::Providers::OpenRouter
 RubyLLM::Provider.register :ollama, RubyLLM::Providers::Ollama
+RubyLLM::Provider.register :mistral, RubyLLM::Providers::Mistral
 
 if defined?(Rails::Railtie)
   require 'ruby_llm/railtie'

lib/ruby_llm/aliases.json

@@ -76,5 +76,14 @@
   "o3-mini": {
     "openai": "o3-mini",
     "openrouter": "openai/o3-mini"
+  },
+  "mistral-medium": {
+    "mistral": "mistral-medium-latest"
+  },
+  "mistral-large": {
+    "mistral": "mistral-large-latest"
+  },
+  "mistral-small": {
+    "mistral": "mistral-small-latest"
   }
 }

lib/ruby_llm/configuration.rb

@@ -22,6 +22,7 @@ class Configuration
                   :bedrock_session_token,
                   :openrouter_api_key,
                   :ollama_api_base,
+                  :mistral_api_key,
                   # Default models
                   :default_model,
                   :default_embedding_model,

lib/ruby_llm/models.json

@@ -11409,6 +11409,24 @@
       }
     }
   },
+  {
+    "id": "mistralai/mistral-embed",
+    "created_at": "2024-03-15T00:00:00Z",
+    "display_name": "Mistral Embed",
+    "provider": "mistral",
+    "context_window": 32768,
+    "max_tokens": 32768,
+    "type": "embedding",
+    "family": "mistral_embed",
+    "supports_vision": false,
+    "supports_functions": false,
+    "supports_json_mode": false,
+    "input_price_per_million": 0.01,
+    "output_price_per_million": 0.01,
+    "metadata": {
+      "description": "Mistral Embed - specialized model for generating text embeddings"
+    }
+  },
   {
     "id": "moonshotai/kimi-vl-a3b-thinking:free",
     "created_at": "2025-04-10T19:07:21+02:00",

lib/ruby_llm/providers/mistral.rb

@@ -0,0 +1,51 @@
+module RubyLLM
+  module Providers
+    # Mistral API integration. Handles chat completion, embeddings,
+    # and Mistral's streaming format. Supports Mistral models.
+    module Mistral
+      extend Provider
+      extend Mistral::Chat
+      extend Mistral::Embeddings
+      extend Mistral::Models
+      extend Mistral::Streaming
+      extend Mistral::Tools
+      extend Mistral::Capabilities
+      extend Mistral::Media
+
+      def self.extended(base)
+        base.extend(Provider)
+        base.extend(Mistral::Chat)
+        base.extend(Mistral::Embeddings)
+        base.extend(Mistral::Models)
+        base.extend(Mistral::Streaming)
+        base.extend(Mistral::Tools)
+        base.extend(Mistral::Capabilities)
+        base.extend(Mistral::Media)
+      end
+
+      module_function
+
+      def api_base
+        'https://api.mistral.ai/v1'
+      end
+
+      def headers
+        {
+          'Authorization' => "Bearer #{RubyLLM.config.mistral_api_key}"
+        }
+      end
+
+      def capabilities
+        Mistral::Capabilities
+      end
+
+      def slug
+        'mistral'
+      end
+
+      def configuration_requirements
+        %i[mistral_api_key]
+      end
+    end
+  end
+end

lib/ruby_llm/providers/mistral/capabilities.rb

@@ -0,0 +1,176 @@
+module RubyLLM
+  module Providers
+    module Mistral
+      # Determines capabilities and pricing for Mistral models
+      module Capabilities
+        module_function
+
+        # Returns the context window size for the given model ID
+        # @param model_id [String] the model identifier
+        # @return [Integer] the context window size in tokens
+        def context_window_for(model_id)
+          case model_id
+          when /mistral-large-2411/ then 131_000
+          when /mistral-small-latest/ then 32_000
+          when /mistral-medium-latest/ then 32_000
+          when /ministral-3b-latest/ then 131_000
+          when /ministral-8b-latest/ then 131_000
+          when /codestral-2501/ then 256_000
+          when /pixtral-large-latest/ then 131_000
+          when /mistral-saba-latest/ then 32_000
+          when /mistral-embed/ then 8_000
+          when /mistral-moderation-latest/ then 8_000
+          when /codestral-mamba-latest/ then 1_000_000 # Using a large number for "Infinite-length"
+          else 32_000 # Default to smallest common size
+          end
+        end
+
+        # Returns the maximum output tokens for the given model ID
+        # @param model_id [String] the model identifier
+        # @return [Integer] the maximum output tokens
+        def max_tokens_for(model_id)
+          # Generally, max output tokens is slightly less than context window
+          (context_window_for(model_id) * 0.9).to_i
+        end
+
+        # Returns the input price per million tokens for the given model ID
+        # @param model_id [String] the model identifier
+        # @return [Float] the price per million tokens for input
+        def input_price_for(model_id)
+          PRICES.dig(model_family(model_id), :input) || default_input_price
+        end
+
+        # Returns the output price per million tokens for the given model ID
+        # @param model_id [String] the model identifier
+        # @return [Float] the price per million tokens for output
+        def output_price_for(model_id)
+          PRICES.dig(model_family(model_id), :output) || default_output_price
+        end
+
+        # Determines if the model supports vision capabilities
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports vision
+        def supports_vision?(model_id)
+          # Check for all vision-capable models in Mistral's lineup
+          #
+          # NOTE: While this correctly identifies models that support vision capabilities,
+          # there are currently issues with the image handling in the test suite.
+          # The 'pixtral-12b-latest can understand images' test fails because image data
+          # isn't properly passed to the API. This requires fixes in the core Content handling code.
+          #
+          # Known vision-capable models in Mistral's lineup:
+          # - pixtral-12b-latest
+          # - pixtral-large-latest
+          model_id.match?(/pixtral|mistral-large-vision|other-vision-models/)
+        end
+
+        # Determines if the model supports function calling
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports functions
+        def supports_functions?(model_id)
+          !model_id.match?(/embed|moderation/)
+        end
+
+        # Determines if the model supports audio input/output
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports audio
+        def supports_audio?(_model_id)
+          false
+        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)
+          !model_id.match?(/embed|moderation/)
+        end
+
+        # Formats the model ID into a human-readable display name
+        # @param model_id [String] the model identifier
+        # @return [String] the formatted display name
+        def format_display_name(model_id)
+          model_id.then { |id| humanize(id) }
+                  .then { |name| apply_special_formatting(name) }
+        end
+
+        # Determines the type of model
+        # @param model_id [String] the model identifier
+        # @return [String] the model type (chat, embedding, moderation)
+        def model_type(model_id)
+          case model_id
+          when /embed/ then "embedding"
+          when /moderation/ then "moderation"
+          else "chat"
+          end
+        end
+
+        # Determines if the model supports structured output
+        # @param model_id [String] the model identifier
+        # @return [Boolean] true if the model supports structured output
+        def supports_structured_output?(model_id)
+          !model_id.match?(/embed|moderation/)
+        end
+
+        # Determines the model family for pricing and capability lookup
+        # @param model_id [String] the model identifier
+        # @return [Symbol] the model family identifier
+        def model_family(model_id)
+          case model_id
+          when /mistral-large/ then "large"
+          when /mistral-small/ then "small"
+          when /codestral/ then "codestral"
+          when /pixtral/ then "pixtral"
+          when /mistral-nemo/ then "nemo"
+          when /embed/ then "embedding"
+          else "other"
+          end
+        end
+
+        # Pricing information for Mistral models (per million tokens)
+        PRICES = {
+          "large" => { input: 2.00, output: 6.00 },
+          "small" => { input: 0.20, output: 0.60 },
+          "codestral" => { input: 0.30, output: 0.90 },
+          "pixtral" => { input: 0.15, output: 0.15 },
+          "nemo" => { input: 0.15, output: 0.15 },
+          "embedding" => { price: 0.1 },
+        }.freeze
+
+        # Default input price when model-specific pricing is not available
+        # @return [Float] the default price per million tokens
+        def default_input_price
+          0.20 # Default to small model pricing
+        end
+
+        # Default output price when model-specific pricing is not available
+        # @return [Float] the default price per million tokens
+        def default_output_price
+          0.60 # Default to small model pricing
+        end
+
+        # Converts a model ID to a human-readable format
+        # @param id [String] the model identifier
+        # @return [String] the humanized model name
+        def humanize(id)
+          id.tr("-", " ")
+            .split
+            .map(&:capitalize)
+            .join(" ")
+        end
+
+        # Applies special formatting rules to model names
+        # @param name [String] the humanized model name
+        # @return [String] the specially formatted model name
+        def apply_special_formatting(name)
+          name
+            .gsub("Mistral ", "Mistral-")
+            .gsub("Ministral ", "Ministral-")
+            .gsub("Codestral ", "Codestral-")
+            .gsub("Pixtral ", "Pixtral-")
+            .gsub("Mathstral ", "Mathstral-")
+            .gsub("Embed ", "Embed-")
+        end
+      end
+    end
+  end
+end