mirascope 1.19.0__py3-none-any.whl → 1.20.0__py3-none-any.whl

mirascope/core/{cohere/_utils/_calculate_cost.py → costs/_cohere_calculate_cost.py}

@@ -1,10 +1,10 @@
 """Calculate the cost of a completion using the Cohere API."""
 
+from ..base.types import CostMetadata
+
 
 def calculate_cost(
-    input_tokens: int | float | None,
-    cached_tokens: int | float | None,
-    output_tokens: int | float | None,
+    metadata: CostMetadata,
     model: str = "command-r-plus",
 ) -> float | None:
     """Calculate the cost of a completion using the Cohere API.
@@ -12,8 +12,8 @@ def calculate_cost(
     https://cohere.com/pricing
 
     Model              Input               Cached     Output
-    command-r          $0.5 / 1M tokens	              $1.5 / 1M tokens
-    command-r-plus     $3 / 1M tokens	              $15 / 1M tokens
+    command-r          $0.15 / 1M tokens              $0.6 / 1M tokens
+    command-r-plus     $2.5 / 1M tokens               $10 / 1M tokens
     """
     pricing = {
         "command-r": {
@@ -24,8 +24,12 @@ def calculate_cost(
             "prompt": 0.000_003,
             "completion": 0.000_015,
         },
+        "command-r7b-12-2024": {
+            "prompt": 0.000_000_375,
+            "completion": 0.000_001_5,
+        },
     }
-    if input_tokens is None or output_tokens is None:
+    if metadata.input_tokens is None or metadata.output_tokens is None:
         return None
 
     try:
@@ -33,8 +37,8 @@ def calculate_cost(
     except KeyError:
         return None
 
-    prompt_cost = input_tokens * model_pricing["prompt"]
-    completion_cost = output_tokens * model_pricing["completion"]
+    prompt_cost = metadata.input_tokens * model_pricing["prompt"]
+    completion_cost = metadata.output_tokens * model_pricing["completion"]
     total_cost = prompt_cost + completion_cost
 
     return total_cost

mirascope/core/{gemini/_utils/_calculate_cost.py → costs/_gemini_calculate_cost.py}

@@ -1,10 +1,10 @@
 """Calculate the cost of a Gemini API call."""
 
+from ..base.types import CostMetadata
+
 
 def calculate_cost(
-    input_tokens: int | float | None,
-    cached_tokens: int | float | None,
-    output_tokens: int | float | None,
+    metadata: CostMetadata,
     model: str,
 ) -> float | None:
     """Calculate the cost of a Gemini API call.
@@ -44,7 +44,7 @@ def calculate_cost(
         },
     }
 
-    if input_tokens is None or output_tokens is None:
+    if metadata.input_tokens is None or metadata.output_tokens is None:
         return None
 
     try:
@@ -53,15 +53,15 @@ def calculate_cost(
         return None
 
     # Determine if we're using long context pricing
-    use_long_context = input_tokens > 128_000
+    use_long_context = metadata.input_tokens > 128_000
 
     prompt_price = model_pricing["prompt_long" if use_long_context else "prompt_short"]
     completion_price = model_pricing[
         "completion_long" if use_long_context else "completion_short"
     ]
 
-    prompt_cost = input_tokens * prompt_price
-    completion_cost = output_tokens * completion_price
+    prompt_cost = metadata.input_tokens * prompt_price
+    completion_cost = metadata.output_tokens * completion_price
     total_cost = prompt_cost + completion_cost
 
     return total_cost

mirascope/core/costs/_google_calculate_cost.py

@@ -0,0 +1,427 @@
+"""Calculate the cost of a Gemini API call."""
+
+from ..base.types import CostMetadata
+
+# Standard Gemini API pricing table
+GEMINI_API_PRICING: dict[str, dict[str, float]] = {
+    "gemini-2.0-pro": {
+        "prompt_short": 0.000_001_25,
+        "completion_short": 0.000_005,
+        "prompt_long": 0.000_002_5,
+        "completion_long": 0.000_01,
+        "cached": 0.000_000_625,
+    },
+    "gemini-2.0-pro-preview-1206": {
+        "prompt_short": 0.000_001_25,
+        "completion_short": 0.000_005,
+        "prompt_long": 0.000_002_5,
+        "completion_long": 0.000_01,
+        "cached": 0.000_000_625,
+    },
+    "gemini-2.0-flash": {
+        "prompt_short": 0.000_000_10,
+        "completion_short": 0.000_000_40,
+        "prompt_long": 0.000_000_10,
+        "completion_long": 0.000_000_40,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-2.0-flash-latest": {
+        "prompt_short": 0.000_000_10,
+        "completion_short": 0.000_000_40,
+        "prompt_long": 0.000_000_10,
+        "completion_long": 0.000_000_40,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-2.0-flash-001": {
+        "prompt_short": 0.000_000_10,
+        "completion_short": 0.000_000_40,
+        "prompt_long": 0.000_000_10,
+        "completion_long": 0.000_000_40,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-2.0-flash-lite": {
+        "prompt_short": 0.000_000_075,
+        "completion_short": 0.000_000_30,
+        "prompt_long": 0.000_000_075,
+        "completion_long": 0.000_000_30,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-2.0-flash-lite-preview-02-05": {
+        "prompt_short": 0.000_000_075,
+        "completion_short": 0.000_000_30,
+        "prompt_long": 0.000_000_075,
+        "completion_long": 0.000_000_30,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-1.5-pro": {
+        "prompt_short": 0.000_001_25,
+        "completion_short": 0.000_005,
+        "prompt_long": 0.000_002_5,
+        "completion_long": 0.000_01,
+        "cached": 0.000_000_625,
+    },
+    "gemini-1.5-pro-latest": {
+        "prompt_short": 0.000_001_25,
+        "completion_short": 0.000_005,
+        "prompt_long": 0.000_002_5,
+        "completion_long": 0.000_01,
+        "cached": 0.000_000_625,
+    },
+    "gemini-1.5-pro-001": {
+        "prompt_short": 0.000_001_25,
+        "completion_short": 0.000_005,
+        "prompt_long": 0.000_002_5,
+        "completion_long": 0.000_01,
+        "cached": 0.000_000_625,
+    },
+    "gemini-1.5-pro-002": {
+        "prompt_short": 0.000_001_25,
+        "completion_short": 0.000_005,
+        "prompt_long": 0.000_002_5,
+        "completion_long": 0.000_01,
+        "cached": 0.000_000_625,
+    },
+    "gemini-1.5-flash": {
+        "prompt_short": 0.000_000_075,
+        "completion_short": 0.000_000_30,
+        "prompt_long": 0.000_000_15,
+        "completion_long": 0.000_000_60,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-1.5-flash-latest": {
+        "prompt_short": 0.000_000_075,
+        "completion_short": 0.000_000_30,
+        "prompt_long": 0.000_000_15,
+        "completion_long": 0.000_000_60,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-1.5-flash-001": {
+        "prompt_short": 0.000_000_075,
+        "completion_short": 0.000_000_30,
+        "prompt_long": 0.000_000_15,
+        "completion_long": 0.000_000_60,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-1.5-flash-002": {
+        "prompt_short": 0.000_000_075,
+        "completion_short": 0.000_000_30,
+        "prompt_long": 0.000_000_15,
+        "completion_long": 0.000_000_60,
+        "cached": 0.000_000_037_5,
+    },
+    "gemini-1.5-flash-8b": {
+        "prompt_short": 0.000_000_037_5,
+        "completion_short": 0.000_000_15,
+        "prompt_long": 0.000_000_075,
+        "completion_long": 0.000_000_30,
+        "cached": 0.000_000_025,
+    },
+    "gemini-1.5-flash-8b-latest": {
+        "prompt_short": 0.000_000_037_5,
+        "completion_short": 0.000_000_15,
+        "prompt_long": 0.000_000_075,
+        "completion_long": 0.000_000_30,
+        "cached": 0.000_000_025,
+    },
+    "gemini-1.5-flash-8b-001": {
+        "prompt_short": 0.000_000_037_5,
+        "completion_short": 0.000_000_15,
+        "prompt_long": 0.000_000_075,
+        "completion_long": 0.000_000_30,
+        "cached": 0.000_000_025,
+    },
+    "gemini-1.5-flash-8b-002": {
+        "prompt_short": 0.000_000_037_5,
+        "completion_short": 0.000_000_15,
+        "prompt_long": 0.000_000_075,
+        "completion_long": 0.000_000_30,
+        "cached": 0.000_000_025,
+    },
+    "gemini-1.0-pro": {
+        "prompt_short": 0.000_000_5,
+        "completion_short": 0.000_001_5,
+        "prompt_long": 0.000_000_5,
+        "completion_long": 0.000_001_5,
+        "cached": 0.000_000,
+    },
+}
+
+# Vertex AI pricing table
+VERTEX_AI_PRICING: dict[str, dict[str, float]] = {
+    "gemini-2.0-flash": {
+        "text_input": 0.000_000_15,
+        "image_input": 0.000_000_15,
+        "video_input": 0.000_000_15,
+        "audio_input": 0.000_001_00,
+        "output": 0.000_000_60,
+        "cached": 0.000_000_037_5,
+        "cache_storage_per_hour": 0.000_001_00,
+    },
+    "gemini-2.0-flash-lite": {
+        "text_input": 0.000_000_075,
+        "image_input": 0.000_000_075,
+        "video_input": 0.000_000_075,
+        "audio_input": 0.000_000_075,
+        "output": 0.000_000_30,
+        "cached": 0.000_000_037_5,
+        "cache_storage_per_hour": 0.000_001_00,
+    },
+    # Vertex AI pricing for Gemini 1.5 models is based on modalities rather than tokens
+    "gemini-1.5-flash": {
+        "text_input": 0.000_000_075,  # per 1K chars (approx. 250 tokens)
+        "image_input": 0.000_02,  # per image
+        "video_input": 0.000_02,  # per second
+        "audio_input": 0.000_002,  # per second
+        "output": 0.000_000_30,  # per 1K chars
+        "cached_text": 0.000_000_046_875,  # per 1K chars
+        "cached_image": 0.000_005,  # per image
+        "cached_video": 0.000_005,  # per second
+        "cached_audio": 0.000_000_5,  # per second
+        "cache_storage_text": 0.000_25,  # per 1K chars per hour
+        "cache_storage_image": 0.000_263,  # per image per hour
+        "cache_storage_video": 0.000_263,  # per second per hour
+        "cache_storage_audio": 0.000_025,  # per second per hour
+    },
+    "gemini-1.5-pro": {
+        "text_input": 0.000_001_25,  # per 1K chars (approx. 250 tokens)
+        "image_input": 0.000_32875,  # per image
+        "video_input": 0.000_32875,  # per second
+        "audio_input": 0.000_03125,  # per second
+        "output": 0.000_005,  # per 1K chars
+        "cached_text": 0.000_000_078125,  # per 1K chars
+        "cached_image": 0.000_0821875,  # per image
+        "cached_video": 0.000_0821875,  # per second
+        "cached_audio": 0.000_0078125,  # per second
+        "cache_storage_text": 0.001125,  # per 1K chars per hour
+        "cache_storage_image": 0.0011835,  # per image per hour
+        "cache_storage_video": 0.0011835,  # per second per hour
+        "cache_storage_audio": 0.0001125,  # per second per hour
+    },
+}
+
+
+def _calculate_context_cache_cost(
+    metadata: CostMetadata,
+    model_pricing: dict[str, float],
+    model: str,
+    use_vertex_ai: bool = False,
+) -> float:
+    """Calculate cost for context caching."""
+    if metadata.context_cache_tokens is None or metadata.context_cache_hours is None:
+        return 0.0
+
+    if use_vertex_ai:
+        # Vertex AI pricing depends on the model family
+        if model.startswith("gemini-2.0"):
+            return (
+                metadata.context_cache_tokens
+                * model_pricing.get("cache_storage_per_hour", 0)
+                * metadata.context_cache_hours
+            )
+        elif model.startswith("gemini-1.5"):
+            # Convert cache tokens to characters (approx)
+            cache_chars = metadata.context_cache_tokens * 4
+            return (
+                (cache_chars / 1000)
+                * model_pricing["cache_storage_text"]
+                * metadata.context_cache_hours
+            )
+
+    # Standard Gemini API pricing - storage cost per token-hour
+    storage_rate_per_token = 0.000001  # $1.00 per million tokens per hour
+    if "flash-8b" in model:
+        storage_rate_per_token = 0.00000025  # $0.25 per million tokens for 8B models
+
+    return (
+        metadata.context_cache_tokens
+        * storage_rate_per_token
+        * metadata.context_cache_hours
+    )
+
+
+def _calculate_grounding_cost(metadata: CostMetadata, model: str) -> float:
+    """Calculate cost for grounding requests."""
+    if metadata.google is None or metadata.google.grounding_requests is None:
+        return 0.0
+
+    # First 1,500 requests per day are free for Gemini 2.0 Flash models in Vertex AI
+    if (
+        model == "gemini-2.0-flash"
+        and metadata.google.use_vertex_ai
+        and metadata.google.grounding_requests <= 1500
+    ):
+        return 0.0
+
+    # $35 per 1,000 requests for excess
+    if metadata.google.use_vertex_ai and model == "gemini-2.0-flash":
+        excess_requests = max(0, metadata.google.grounding_requests - 1500)
+    else:
+        excess_requests = metadata.google.grounding_requests
+
+    return (excess_requests / 1000) * 35.0
+
+
+def _calculate_vertex_2_0_cost(
+    metadata: CostMetadata, model_pricing: dict[str, float], model: str
+) -> float:
+    """Calculate cost for Vertex AI's Gemini 2.0 models."""
+    # Text tokens cost
+    prompt_cost = (metadata.input_tokens or 0) * model_pricing["text_input"]
+    completion_cost = (metadata.output_tokens or 0) * model_pricing["output"]
+    cached_cost = (metadata.cached_tokens or 0) * model_pricing.get("cached", 0)
+
+    # Context cache costs
+    context_cache_cost = _calculate_context_cache_cost(
+        metadata, model_pricing, model, use_vertex_ai=True
+    )
+
+    # Grounding costs
+    grounding_cost = _calculate_grounding_cost(metadata, model)
+
+    # Apply batch mode discount (50% for Vertex AI)
+    if metadata.batch_mode:
+        prompt_cost *= 0.5
+        completion_cost *= 0.5
+        context_cache_cost *= 0.5
+        # Note: We don't discount grounding costs
+
+    total_cost = (
+        prompt_cost
+        + completion_cost
+        + cached_cost
+        + context_cache_cost
+        + grounding_cost
+    )
+
+    return total_cost
+
+
+def _calculate_vertex_1_5_cost(
+    metadata: CostMetadata, model_pricing: dict[str, float], model: str
+) -> float:
+    """Calculate cost for Vertex AI's Gemini 1.5 models."""
+    # Text cost - convert tokens to characters (approx 4 chars per token)
+    text_chars = (metadata.input_tokens or 0) * 4  # Approximation
+    text_cost = (text_chars / 1000) * model_pricing["text_input"]
+
+    # Output cost
+    output_chars = (metadata.output_tokens or 0) * 4  # Approximation
+    output_cost = (output_chars / 1000) * model_pricing["output"]
+
+    # Context cache costs
+    context_cache_cost = _calculate_context_cache_cost(
+        metadata, model_pricing, model, use_vertex_ai=True
+    )
+
+    # Grounding costs
+    grounding_cost = _calculate_grounding_cost(metadata, model)
+
+    # Apply batch mode discount if applicable (50% off for Vertex AI)
+    if metadata.batch_mode:
+        text_cost *= 0.5
+        output_cost *= 0.5
+        context_cache_cost *= 0.5
+        # Note: We don't discount grounding costs
+
+    total_cost = text_cost + output_cost + context_cache_cost + grounding_cost
+
+    return total_cost
+
+
+def _calculate_standard_gemini_cost(
+    metadata: CostMetadata,
+    model_pricing: dict[str, float],
+    model: str,
+    use_long_context: bool,
+) -> float:
+    """Calculate cost for standard Gemini API."""
+    # Determine prices based on context length
+    prompt_price = model_pricing["prompt_long" if use_long_context else "prompt_short"]
+    cached_price = model_pricing["cached"]
+    completion_price = model_pricing[
+        "completion_long" if use_long_context else "completion_short"
+    ]
+
+    # Basic token costs
+    prompt_cost = (metadata.input_tokens or 0) * prompt_price
+    cached_cost = (metadata.cached_tokens or 0) * cached_price
+    completion_cost = (metadata.output_tokens or 0) * completion_price
+
+    # Media token costs is included in the prompt/completion cost
+
+    # Context cache costs
+    context_cache_cost = _calculate_context_cache_cost(
+        metadata, model_pricing, model, use_vertex_ai=False
+    )
+
+    # Grounding costs - only applies to certain models
+    grounding_cost = _calculate_grounding_cost(metadata, model)
+
+    total_cost = (
+        prompt_cost
+        + cached_cost
+        + completion_cost
+        + context_cache_cost
+        + grounding_cost
+    )
+
+    return total_cost
+
+
+def calculate_cost(
+    metadata: CostMetadata,
+    model: str,
+) -> float | None:
+    """Calculate the cost of a Google API call.
+
+    This function supports both direct Gemini API and Vertex AI pricing.
+    It handles different media types (text, image, video, audio) and special features
+    like context caching and grounding.
+
+    https://ai.google.dev/pricing
+    https://cloud.google.com/vertex-ai/generative-ai/pricing
+
+    Args:
+        metadata: Additional metadata required for cost calculation
+        model: Model name to use for pricing calculation
+
+    Returns:
+        Total cost in USD or None if invalid input
+    """
+    # Basic validation
+    if metadata.input_tokens is None or metadata.output_tokens is None:
+        return None
+
+    # Initialize default values
+    if metadata.cached_tokens is None:
+        metadata.cached_tokens = 0
+
+    # Check if we're using Vertex AI pricing
+    use_vertex_ai = metadata.google and metadata.google.use_vertex_ai
+
+    # Determine if we're using long context pricing
+    use_long_context = (
+        metadata.context_length is not None and metadata.context_length > 128_000
+    ) or (metadata.input_tokens > 128_000)
+
+    # Get the appropriate pricing table
+    try:
+        if use_vertex_ai and model in VERTEX_AI_PRICING:
+            model_pricing = VERTEX_AI_PRICING[model]
+        else:
+            model_pricing = GEMINI_API_PRICING[model]
+    except KeyError:
+        # Unknown model
+        return None
+
+    # Calculate cost based on API type
+    if use_vertex_ai:
+        if model.startswith("gemini-2.0"):
+            return _calculate_vertex_2_0_cost(metadata, model_pricing, model)
+        elif model.startswith("gemini-1.5"):  # pragma: no cover
+            return _calculate_vertex_1_5_cost(metadata, model_pricing, model)
+    else:
+        # Standard Gemini API pricing
+        return _calculate_standard_gemini_cost(
+            metadata, model_pricing, model, use_long_context
+        )

mirascope/core/costs/_groq_calculate_cost.py

@@ -0,0 +1,156 @@
+"""Calculate the cost of a completion using the Groq API."""
+
+from ..base.types import CostMetadata
+
+
+def calculate_cost(
+    metadata: CostMetadata,
+    model: str = "mixtral-8x7b-32768",
+) -> float | None:
+    """Calculate the cost of a completion using the Groq API.
+
+    https://wow.groq.com/
+
+    Model                                  Input                  Output
+    llama-3.3-70b-versatile                $0.59 / 1M tokens      $0.79 / 1M tokens
+    llama-3.3-70b-specdec                  $0.59 / 1M tokens      $0.99 / 1M tokens
+    llama-3.1-8b-instant                   $0.05 / 1M tokens      $0.08 / 1M tokens
+    llama3-70b-8192                        $0.59 / 1M tokens      $0.79 / 1M tokens
+    llama3-8b-8192                         $0.05 / 1M tokens      $0.08 / 1M tokens
+    llama-guard-3-8b                       $0.20 / 1M tokens      $0.20 / 1M tokens
+    mixtral-8x7b-32768                     $0.24 / 1M tokens      $0.24 / 1M tokens
+    gemma-7b-it                            $0.07 / 1M tokens      $0.07 / 1M tokens
+    gemma2-9b-it                           $0.20 / 1M tokens      $0.20 / 1M tokens
+    mistral-saba-24b                       $0.79 / 1M tokens      $0.79 / 1M tokens
+    qwen-2.5-32b                           $0.79 / 1M tokens      $0.79 / 1M tokens
+    qwen-2.5-coder-32b                     $0.79 / 1M tokens      $0.79 / 1M tokens
+    deepseek-r1-distill-qwen-32b           $0.69 / 1M tokens      $0.69 / 1M tokens
+    deepseek-r1-distill-llama-70b          $0.75 / 1M tokens      $0.99 / 1M tokens
+    deepseek-r1-distill-llama-70b-specdec  $0.75 / 1M tokens      $0.99 / 1M tokens
+    llama-3.2-1b-preview                   $0.04 / 1M tokens      $0.04 / 1M tokens
+    llama-3.2-3b-preview                   $0.06 / 1M tokens      $0.06 / 1M tokens
+    llama-3.2-11b-vision-preview           $0.18 / 1M tokens      $0.18 / 1M tokens
+    llama-3.2-90b-vision-preview           $0.90 / 1M tokens      $0.90 / 1M tokens
+    """
+    pricing = {
+        "llama-3.3-70b-versatile": {
+            "prompt": 0.000_000_59,
+            "completion": 0.000_000_79,
+        },
+        "llama-3.3-70b-specdec": {
+            "prompt": 0.000_000_59,
+            "completion": 0.000_000_99,
+        },
+        "llama3-groq-70b-8192-tool-use-preview": {
+            "prompt": 0.000_000_89,
+            "completion": 0.000_000_89,
+        },
+        "llama3-groq-8b-8192-tool-use-preview": {
+            "prompt": 0.000_000_19,
+            "completion": 0.000_000_19,
+        },
+        "llama-3.1-8b-instant": {
+            "prompt": 0.000_000_05,
+            "completion": 0.000_000_08,
+        },
+        "llama-guard-3-8b": {
+            "prompt": 0.000_000_2,
+            "completion": 0.000_000_2,
+        },
+        "llama3-70b-8192": {
+            "prompt": 0.000_000_59,
+            "completion": 0.000_000_79,
+        },
+        "llama3-8b-8192": {
+            "prompt": 0.000_000_05,
+            "completion": 0.000_000_08,
+        },
+        "mixtral-8x7b-32768": {
+            "prompt": 0.000_000_24,
+            "completion": 0.000_000_24,
+        },
+        "gemma-7b-it": {
+            "prompt": 0.000_000_07,
+            "completion": 0.000_000_07,
+        },
+        "gemma2-9b-it": {
+            "prompt": 0.000_000_2,
+            "completion": 0.000_000_2,
+        },
+        "mistral-saba-24b": {
+            "prompt": 0.000_000_79,
+            "completion": 0.000_000_79,
+        },
+        "qwen-2.5-32b": {
+            "prompt": 0.000_000_79,
+            "completion": 0.000_000_79,
+        },
+        "qwen-2.5-coder-32b": {
+            "prompt": 0.000_000_79,
+            "completion": 0.000_000_79,
+        },
+        "deepseek-r1-distill-qwen-32b": {
+            "prompt": 0.000_000_69,
+            "completion": 0.000_000_69,
+        },
+        "deepseek-r1-distill-llama-70b": {
+            "prompt": 0.000_000_75,
+            "completion": 0.000_000_99,
+        },
+        "deepseek-r1-distill-llama-70b-specdec": {
+            "prompt": 0.000_000_75,
+            "completion": 0.000_000_99,
+        },
+        "llama-3.2-1b-preview": {
+            "prompt": 0.000_000_04,
+            "completion": 0.000_000_04,
+        },
+        "llama-3.2-3b-preview": {
+            "prompt": 0.000_000_06,
+            "completion": 0.000_000_06,
+        },
+        # Vision models
+        "llama-3.2-11b-vision-preview": {
+            "prompt": 0.000_000_18,
+            "completion": 0.000_000_18,
+        },
+        "llama-3.2-90b-vision-preview": {
+            "prompt": 0.000_000_90,
+            "completion": 0.000_000_90,
+        },
+    }
+
+    if metadata.input_tokens is None or metadata.output_tokens is None:
+        return None
+
+    try:
+        model_pricing = pricing[model]
+    except KeyError:
+        return None
+
+    # Calculate cost for text tokens
+    prompt_cost = metadata.input_tokens * model_pricing["prompt"]
+    completion_cost = metadata.output_tokens * model_pricing["completion"]
+    total_cost = prompt_cost + completion_cost
+
+    # Calculate cost for images if present
+    # Groq bills 6,400 tokens per image for vision models
+    # https://groq.com/pricing/
+    image_cost = 0.0
+    if metadata.images and "vision" in model:
+        # For Groq vision models, each image is billed at 6,400 tokens
+        tokens_per_image = 6400
+
+        # Count the number of images
+        image_count = len(metadata.images)
+
+        # Calculate total image tokens
+        total_image_tokens = image_count * tokens_per_image
+
+        # Images are charged at the prompt token rate
+        image_cost = total_image_tokens * model_pricing["prompt"]
+
+    # Add image cost to total
+    total_cost += image_cost
+
+    return total_cost

mirascope/core/costs/_litellm_calculate_cost.py

@@ -0,0 +1,11 @@
+"""Calculate the cost of a Litellm call."""
+
+from ..base.types import CostMetadata
+
+
+def calculate_cost(
+    metadata: CostMetadata,
+    model: str,
+) -> float | None:
+    """Calculate the cost of a Litellm call."""
+    return metadata.cost