agentbyte 0.56.0__tar.gz → 0.57.0__tar.gz

{agentbyte-0.56.0 → agentbyte-0.57.0}/CHANGELOG.md

@@ -6,6 +6,24 @@ The format follows Keep a Changelog principles and semantic versioning.
 
 ## [Unreleased]
 
+## [0.57.0] - 2026-10-27
+
+### Changed
+
+- **Breaking:** manual pricing is now `pricing_override=` on all chat and embedding clients and
+  factories (formerly `model_pricing=`). `model` is the invocation target, `pricing_model` selects
+  catalog pricing, and `pricing_override` supplies explicit rates (spec 0078).
+- Pricing lookup is shared across providers: exact, normalized (Bedrock inference-profile IDs),
+  boundary-compatible prefix/family, then default, with warnings for approximate and default matches.
+  `eu.cohere.embed-v4:0` now resolves to the Cohere v4 entry.
+
+### Fixed
+
+- Anthropic and Bedrock Anthropic cost estimates no longer go negative with prompt caching.
+  Anthropic's `input_tokens` excludes cache reads, so `Usage.tokens_input` now includes
+  `cache_read_input_tokens` and `cache_creation_input_tokens` (normal and streaming), and
+  `tokens_cached` stays a subset of it. Cache-write tokens are billed at the plain input rate.
+
 ## [0.56.0] - 2026-10-23
 
 ### Added

{agentbyte-0.56.0 → agentbyte-0.57.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: agentbyte
-Version: 0.56.0
+Version: 0.57.0
 Summary: A toolkit for designing multiagent systems
 Author-email: MrDataPsycho <mr.data.psycho@gmail.com>
 License-Expression: LicenseRef-Proprietary
@@ -90,7 +90,7 @@ Description-Content-Type: text/markdown
 
 Agentbyte is an observability-first agentic AI framework for building and studying multiagent systems with a learning-first, implementation-oriented workflow.
 
-Current release: **0.56.0**
+Current release: **0.57.0**
 
 ## Building an Agent
 
@@ -323,6 +323,10 @@ Examples are grouped by capability. Start with the folder indexes for
 [optimization](examples/optim/), [evaluation](examples/eval/), or
 [orchestration](examples/orchestration/).
 
+For deployment aliases and custom rates across chat and embedding providers, see the
+[pricing guide](docs/design-guide/01-agent_topics/topic_model_client.md#catalog-pricing-and-deployment-aliases)
+and [offline pricing example](examples/llm/pricing_resolution.py).
+
 ## Installation
 
 Python requirement: **3.11+**

{agentbyte-0.56.0 → agentbyte-0.57.0}/README.md

@@ -2,7 +2,7 @@
 
 Agentbyte is an observability-first agentic AI framework for building and studying multiagent systems with a learning-first, implementation-oriented workflow.
 
-Current release: **0.56.0**
+Current release: **0.57.0**
 
 ## Building an Agent
 
@@ -235,6 +235,10 @@ Examples are grouped by capability. Start with the folder indexes for
 [optimization](examples/optim/), [evaluation](examples/eval/), or
 [orchestration](examples/orchestration/).
 
+For deployment aliases and custom rates across chat and embedding providers, see the
+[pricing guide](docs/design-guide/01-agent_topics/topic_model_client.md#catalog-pricing-and-deployment-aliases)
+and [offline pricing example](examples/llm/pricing_resolution.py).
+
 ## Installation
 
 Python requirement: **3.11+**

agentbyte-0.57.0/src/agentbyte/__about__.py

@@ -0,0 +1,2 @@
+__version__ = "0.57.0"
+VERSION = __version__

{agentbyte-0.56.0 → agentbyte-0.57.0}/src/agentbyte/llm/anthropic/_conversion.py

@@ -126,7 +126,7 @@ class AnthropicConversionMixin:
             except Exception as e:
                 logger.warning(f"Failed to parse structured output: {e}. Using raw content.")
 
-        cached_tokens = response.usage.cache_read_input_tokens or 0
+        input_tokens, cached_tokens = self._normalize_input_usage(response.usage)
 
         return ChatCompletionResult(
             message=AssistantMessage(
@@ -137,12 +137,12 @@ class AnthropicConversionMixin:
             usage=Usage(
                 duration_ms=duration_ms,
                 llm_calls=1,
-                tokens_input=response.usage.input_tokens,
+                tokens_input=input_tokens,
                 tokens_output=response.usage.output_tokens,
                 tokens_cached=cached_tokens,
                 tool_calls=len(tool_calls) if tool_calls else 0,
                 cost_estimate=self._estimate_cost(
-                    response.usage.input_tokens,
+                    input_tokens,
                     response.usage.output_tokens,
                     cached_tokens,
                 ),
@@ -214,6 +214,9 @@ class AnthropicConversionMixin:
         )
 
         model_name = self.model
+        # Input/cache counts arrive on message_start; message_delta may repeat
+        # them (cumulative) or omit them, so keep the latest non-empty values.
+        input_tokens, cached_tokens = 0, 0
         # Anthropic's input_json_delta streams raw JSON-fragment text per
         # content-block index, the same accumulator shape openai/chat.py uses
         # for streamed tool-call arguments.
@@ -222,6 +225,9 @@ class AnthropicConversionMixin:
         async for event in stream:
             if event.type == "message_start":
                 model_name = event.message.model
+                start_usage = getattr(event.message, "usage", None)
+                if start_usage is not None:
+                    input_tokens, cached_tokens = self._normalize_input_usage(start_usage)
                 continue
 
             if event.type == "content_block_start":
@@ -266,9 +272,10 @@ class AnthropicConversionMixin:
 
             if event.type == "message_delta":
                 duration_ms = int((time.perf_counter() - start_time) * 1000)
-                input_tokens = event.usage.input_tokens or 0
+                delta_input, delta_cached = self._normalize_input_usage(event.usage)
+                if delta_input:
+                    input_tokens, cached_tokens = delta_input, delta_cached
                 output_tokens = event.usage.output_tokens or 0
-                cached_tokens = event.usage.cache_read_input_tokens or 0
                 usage = Usage(
                     duration_ms=duration_ms,
                     llm_calls=1,
@@ -287,6 +294,20 @@ class AnthropicConversionMixin:
                 )
                 continue
 
+    @staticmethod
+    def _normalize_input_usage(usage: Any) -> tuple[int, int]:
+        """Return `(total_input_tokens, cached_tokens)` from an Anthropic usage block.
+
+        Anthropic reports `input_tokens` as only the uncached tokens after the
+        last cache breakpoint; cache reads and cache writes are separate
+        counters. AgentByte's `Usage.tokens_input` is the total prompt size with
+        `tokens_cached` as a subset, so fold the counters together here.
+        """
+        uncached = getattr(usage, "input_tokens", None) or 0
+        cache_read = getattr(usage, "cache_read_input_tokens", None) or 0
+        cache_write = getattr(usage, "cache_creation_input_tokens", None) or 0
+        return uncached + cache_read + cache_write, cache_read
+
     @staticmethod
     def _extract_system(messages: List[Message]) -> Optional[str]:
         """Pull SystemMessages out as Anthropic's top-level `system` param."""

{agentbyte-0.56.0 → agentbyte-0.57.0}/src/agentbyte/llm/anthropic/chat.py

@@ -10,7 +10,7 @@ from agentbyte.component import Component
 from agentbyte.middleware.base import ClientMiddleware
 
 from ..base import BaseChatCompletionClient, BaseChatCompletionClientConfig
-from ..pricing import PricingRegistry
+from ..pricing import PricingModelName, PricingRegistry, validate_pricing_model
 from ..retry_policy import RetryMixin, RetryPolicy
 from ..types import ModelPricing
 from ._conversion import AnthropicConversionMixin
@@ -19,7 +19,7 @@ from ._conversion import AnthropicConversionMixin
 class AnthropicChatCompletionClientConfig(BaseChatCompletionClientConfig):
     """Configuration for AnthropicChatCompletionClient serialization."""
 
-    pricing_model: str | None = None
+    pricing_model: PricingModelName | None = None
     retry_policy: RetryPolicy = RetryPolicy()
     key_env_var: str = "ANTHROPIC_API_KEY"
 
@@ -41,7 +41,7 @@ class AnthropicChatCompletionClient(
         model: str,
         client: AsyncAnthropic,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -50,8 +50,8 @@ class AnthropicChatCompletionClient(
         pricing_model: str | None = None,
     ):
         super().__init__(model, client, config, middlewares=middlewares)
-        self.model_pricing = model_pricing
-        self.pricing_model = pricing_model
+        self.pricing_override = pricing_override
+        self.pricing_model = validate_pricing_model(pricing_model)
         if retry_policy is not None:
             self.retry_policy = retry_policy
         else:
@@ -80,7 +80,7 @@ class AnthropicChatCompletionClient(
         api_key: Optional[str] = None,
         base_url: Optional[str] = None,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -112,7 +112,7 @@ class AnthropicChatCompletionClient(
             model=model,
             client=anthropic_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -128,10 +128,12 @@ class AnthropicChatCompletionClient(
         cached_tokens: int = 0,
     ) -> float:
         """Estimate cost from token usage, discounting cached prompt tokens."""
-        if self.model_pricing is not None:
-            pricing = self.model_pricing
+        if self.pricing_override is not None:
+            pricing = self.pricing_override
         else:
-            pricing = PricingRegistry.resolve_anthropic_chat_pricing(self.pricing_model or self.model)
+            pricing = PricingRegistry.resolve_anthropic_chat_pricing(
+                self.pricing_model if self.pricing_model is not None else self.model
+            )
         return pricing.calculate_cost(input_tokens, output_tokens, cached_tokens)
 
     def _to_config(self) -> AnthropicChatCompletionClientConfig:

{agentbyte-0.56.0 → agentbyte-0.57.0}/src/agentbyte/llm/azure/chat.py

@@ -24,7 +24,7 @@ from ..base import (
     AuthenticationError,
     InvalidRequestError,
 )
-from ..pricing import PricingRegistry
+from ..pricing import PricingModelName, PricingRegistry, validate_pricing_model
 from ..retry_policy import RetryMixin, RetryPolicy
 from ..types import (
     ChatCompletionChunk,
@@ -43,7 +43,7 @@ class AzureOpenAIChatCompletionClientConfig(BaseChatCompletionClientConfig):
     """Configuration for AzureOpenAIChatCompletionClient serialization."""
 
     retry_policy: RetryPolicy = RetryPolicy()
-    pricing_model: str | None = None
+    pricing_model: PricingModelName | None = None
     key_env_var: str = "AZURE_OPENAI_API_KEY"
     deployment_env_var: str = "AZURE_OPENAI_CHAT_DEPLOYMENT_NAME"
     api_version_env_var: str = "AZURE_OPENAI_API_VERSION"
@@ -65,7 +65,7 @@ class AzureOpenAIChatCompletionClient(
         model: str,
         client: AsyncAzureOpenAI,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         pricing_model: str | None = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
@@ -74,8 +74,8 @@ class AzureOpenAIChatCompletionClient(
         middlewares: Optional[List[ClientMiddleware]] = None,
     ):
         super().__init__(model, client, config, middlewares=middlewares)
-        self.model_pricing = model_pricing
-        self.pricing_model = pricing_model
+        self.pricing_override = pricing_override
+        self.pricing_model = validate_pricing_model(pricing_model)
         if retry_policy is not None:
             self.retry_policy = retry_policy
         else:
@@ -106,7 +106,7 @@ class AzureOpenAIChatCompletionClient(
         api_key: Optional[str] = None,
         service_settings: Optional[Any] = None,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -156,7 +156,7 @@ class AzureOpenAIChatCompletionClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -174,7 +174,7 @@ class AzureOpenAIChatCompletionClient(
         service_settings: Optional[Any] = None,
         scope: str = "https://cognitiveservices.azure.com/.default",
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -225,7 +225,7 @@ class AzureOpenAIChatCompletionClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -244,7 +244,7 @@ class AzureOpenAIChatCompletionClient(
         principal_settings: Optional[Any] = None,
         scope: Optional[str] = None,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -304,7 +304,7 @@ class AzureOpenAIChatCompletionClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -322,7 +322,7 @@ class AzureOpenAIChatCompletionClient(
         api_version: Optional[str] = None,
         service_settings: Optional[Any] = None,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -365,7 +365,7 @@ class AzureOpenAIChatCompletionClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -662,11 +662,11 @@ class AzureOpenAIChatCompletionClient(
         cached_tokens: int = 0,
     ) -> float:
         """Estimate cost from token usage, discounting cached prompt tokens."""
-        if self.model_pricing is not None:
-            pricing = self.model_pricing
+        if self.pricing_override is not None:
+            pricing = self.pricing_override
         else:
             pricing = PricingRegistry.resolve_azure_chat_pricing(
-                self.pricing_model or self.model
+                self.pricing_model if self.pricing_model is not None else self.model
             )
         return pricing.calculate_cost(prompt_tokens, completion_tokens, cached_tokens)
 

{agentbyte-0.56.0 → agentbyte-0.57.0}/src/agentbyte/llm/azure/embedding.py

@@ -15,7 +15,7 @@ from agentbyte.middleware.base import ClientMiddleware
 from .._retry_observability import build_retry_summary
 from ..base import AuthenticationError, InvalidRequestError, RateLimitError
 from ..embeddings_base import BaseEmbeddingClient, BaseEmbeddingClientConfig
-from ..pricing import PricingRegistry
+from ..pricing import PricingModelName, PricingRegistry, validate_pricing_model
 from ..retry_policy import RetryMixin, RetryPolicy
 from ..types import EmbeddingResult, ModelClientError, ModelPricing, Usage
 
@@ -26,7 +26,7 @@ class AzureOpenAIEmbeddingClientConfig(BaseEmbeddingClientConfig):
     """Configuration for AzureOpenAIEmbeddingClient serialization."""
 
     retry_policy: RetryPolicy = RetryPolicy()
-    pricing_model: str | None = None
+    pricing_model: PricingModelName | None = None
     key_env_var: str = "AZURE_OPENAI_API_KEY"
     deployment_env_var: str = "AZURE_OPENAI_EMBEDDING_DEPLOYMENT_NAME"
     api_version_env_var: str = "AZURE_OPENAI_API_VERSION"
@@ -48,7 +48,7 @@ class AzureOpenAIEmbeddingClient(
         model: str,
         client: AsyncAzureOpenAI,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         pricing_model: str | None = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
@@ -57,8 +57,8 @@ class AzureOpenAIEmbeddingClient(
         middlewares: Optional[List[ClientMiddleware]] = None,
     ):
         super().__init__(model, client, config, middlewares=middlewares)
-        self.model_pricing = model_pricing
-        self.pricing_model = pricing_model
+        self.pricing_override = pricing_override
+        self.pricing_model = validate_pricing_model(pricing_model)
         if retry_policy is not None:
             self.retry_policy = retry_policy
         else:
@@ -90,7 +90,7 @@ class AzureOpenAIEmbeddingClient(
         service_settings: Optional[Any] = None,
         config: Optional[Dict[str, Any]] = None,
 
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -140,7 +140,7 @@ class AzureOpenAIEmbeddingClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -159,7 +159,7 @@ class AzureOpenAIEmbeddingClient(
         scope: str = "https://cognitiveservices.azure.com/.default",
         config: Optional[Dict[str, Any]] = None,
 
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -210,7 +210,7 @@ class AzureOpenAIEmbeddingClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -230,7 +230,7 @@ class AzureOpenAIEmbeddingClient(
         scope: Optional[str] = None,
         config: Optional[Dict[str, Any]] = None,
 
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -290,7 +290,7 @@ class AzureOpenAIEmbeddingClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -309,7 +309,7 @@ class AzureOpenAIEmbeddingClient(
         service_settings: Optional[Any] = None,
         config: Optional[Dict[str, Any]] = None,
 
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -352,7 +352,7 @@ class AzureOpenAIEmbeddingClient(
             model=resolved_model,
             client=azure_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -468,11 +468,11 @@ class AzureOpenAIEmbeddingClient(
 
     def _estimate_cost(self, prompt_tokens: int, completion_tokens: int = 0) -> float:
         """Estimate embedding call cost from token usage."""
-        if self.model_pricing is not None:
-            pricing = self.model_pricing
+        if self.pricing_override is not None:
+            pricing = self.pricing_override
         else:
             pricing = PricingRegistry.resolve_azure_embedding_pricing(
-                self.pricing_model or self.model
+                self.pricing_model if self.pricing_model is not None else self.model
             )
         return pricing.calculate_cost(prompt_tokens, completion_tokens)
 

{agentbyte-0.56.0 → agentbyte-0.57.0}/src/agentbyte/llm/bedrock/anthropic_chat.py

@@ -1,6 +1,5 @@
 """Anthropic Claude chat client transported through AWS Bedrock."""
 
-import re
 from typing import Any, Dict, List, Optional
 
 from anthropic import AsyncAnthropicBedrock
@@ -9,25 +8,17 @@ from agentbyte.component import Component
 from agentbyte.middleware.base import ClientMiddleware
 from agentbyte.llm.anthropic._conversion import AnthropicConversionMixin
 from agentbyte.llm.base import BaseChatCompletionClient, BaseChatCompletionClientConfig
-from agentbyte.llm.pricing import PricingRegistry
+from agentbyte.llm.pricing import PricingModelName, PricingRegistry, validate_pricing_model
 from agentbyte.llm.retry_policy import RetryMixin, RetryPolicy
 from agentbyte.llm.types import ModelPricing
 
 from .settings import BedrockSettings
 
-# Bedrock model/inference-profile IDs carry an optional region prefix ("eu."/
-# "global."), an "anthropic." prefix, and a trailing "-vN:0" suffix (e.g.
-# "eu.anthropic.claude-opus-5-5-v1:0"), none of which the shared Anthropic
-# pricing table's longest-prefix match understands. Strip all three before
-# resolving.
-_BEDROCK_ID_REGION_PREFIX_RE = re.compile(r"^[a-z]+\.(?=anthropic\.)")
-_BEDROCK_ID_SUFFIX_RE = re.compile(r"-v\d+:\d+$")
-
 
 class AnthropicBedrockChatCompletionClientConfig(BaseChatCompletionClientConfig):
     """Serializable Bedrock Claude client configuration."""
 
-    pricing_model: str | None = None
+    pricing_model: PricingModelName | None = None
     retry_policy: RetryPolicy = RetryPolicy()
     region_env_var: str = "AWS_REGION"
     profile_env_var: str = "AWS_PROFILE"
@@ -64,7 +55,7 @@ class AnthropicBedrockChatCompletionClient(
         model: str,
         client: AsyncAnthropicBedrock,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -73,8 +64,8 @@ class AnthropicBedrockChatCompletionClient(
         pricing_model: str | None = None,
     ):
         super().__init__(model, client, config, middlewares=middlewares)
-        self.model_pricing = model_pricing
-        self.pricing_model = pricing_model
+        self.pricing_override = pricing_override
+        self.pricing_model = validate_pricing_model(pricing_model)
         if retry_policy is not None:
             self.retry_policy = retry_policy
         else:
@@ -103,7 +94,7 @@ class AnthropicBedrockChatCompletionClient(
         region: Optional[str] = None,
         profile: Optional[str] = None,
         config: Optional[Dict[str, Any]] = None,
-        model_pricing: Optional[ModelPricing] = None,
+        pricing_override: Optional[ModelPricing] = None,
         max_retries: int = 3,
         initial_retry_delay: float = 1.0,
         max_retry_delay: float = 60.0,
@@ -130,7 +121,7 @@ class AnthropicBedrockChatCompletionClient(
             model=model,
             client=bedrock_client,
             config=config,
-            model_pricing=model_pricing,
+            pricing_override=pricing_override,
             pricing_model=pricing_model,
             max_retries=max_retries,
             initial_retry_delay=initial_retry_delay,
@@ -146,16 +137,12 @@ class AnthropicBedrockChatCompletionClient(
         cached_tokens: int = 0,
     ) -> float:
         """Estimate cost, normalizing the Bedrock model ID before resolving."""
-        if self.model_pricing is not None:
-            pricing = self.model_pricing
-        elif self.pricing_model is not None:
-            pricing = PricingRegistry.resolve_anthropic_chat_pricing(self.pricing_model)
+        if self.pricing_override is not None:
+            pricing = self.pricing_override
         else:
-            stripped_region = _BEDROCK_ID_REGION_PREFIX_RE.sub("", self.model)
-            normalized_model = _BEDROCK_ID_SUFFIX_RE.sub(
-                "", stripped_region.removeprefix("anthropic.")
+            pricing = PricingRegistry.resolve_anthropic_chat_pricing(
+                self.pricing_model if self.pricing_model is not None else self.model
             )
-            pricing = PricingRegistry.resolve_anthropic_chat_pricing(normalized_model)
         return pricing.calculate_cost(input_tokens, output_tokens, cached_tokens)
 
     def _to_config(self) -> AnthropicBedrockChatCompletionClientConfig:

{agentbyte-0.56.0 → agentbyte-0.57.0}/src/agentbyte/llm/bedrock/cohere_embedding.py

@@ -10,7 +10,7 @@ from agentbyte.llm._retry_observability import build_retry_summary
 from agentbyte.middleware.base import ClientMiddleware
 from agentbyte.llm.base import AuthenticationError, InvalidRequestError, RateLimitError
 from agentbyte.llm.embeddings_base import BaseEmbeddingClient, BaseEmbeddingClientConfig
-from agentbyte.llm.pricing import PricingRegistry
+from agentbyte.llm.pricing import PricingModelName, PricingRegistry, validate_pricing_model
 from agentbyte.llm.retry_policy import RetryMixin, RetryPolicy
 from agentbyte.llm.types import EmbeddingResult, ModelClientError, ModelPricing, Usage
 
@@ -32,7 +32,7 @@ class CohereTruncate(StrEnum):
 
 
 class CohereBedrockEmbeddingClientConfig(BaseEmbeddingClientConfig):
-    pricing_model: str | None = None
+    pricing_model: PricingModelName | None = None
     retry_policy: RetryPolicy = RetryPolicy()
     region_env_var: str = "AWS_REGION"
     profile_env_var: str = "AWS_PROFILE"
@@ -48,25 +48,25 @@ class CohereBedrockEmbeddingClient(
     component_provider_override = "agentbyte.llm.CohereBedrockEmbeddingClient"
 
     def __init__(self, model: str, client: Any | None = None, config: Optional[Dict[str, Any]] = None,
-                 model_pricing: Optional[ModelPricing] = None, max_retries: int = 3,
+                 pricing_override: Optional[ModelPricing] = None, max_retries: int = 3,
                  initial_retry_delay: float = 1.0, max_retry_delay: float = 60.0,
                  retry_policy: Optional[RetryPolicy] = None, region: str | None = None,
                  profile: str | None = None, middlewares: Optional[List[ClientMiddleware]] = None,
                  pricing_model: str | None = None):
         super().__init__(model, client, config, middlewares=middlewares)
-        self.pricing_model = pricing_model
-        self.model_pricing, self.region, self.profile = model_pricing, region, profile
+        self.pricing_model = validate_pricing_model(pricing_model)
+        self.pricing_override, self.region, self.profile = pricing_override, region, profile
         self.retry_policy = retry_policy or RetryPolicy(max_retries=max_retries, initial_retry_delay=initial_retry_delay, max_retry_delay=max_retry_delay)
 
     @classmethod
     def from_region(cls, model: str, region: str | None = None, profile: str | None = None,
-                    config: Optional[Dict[str, Any]] = None, model_pricing: Optional[ModelPricing] = None,
+                    config: Optional[Dict[str, Any]] = None, pricing_override: Optional[ModelPricing] = None,
                     max_retries: int = 3, initial_retry_delay: float = 1.0, max_retry_delay: float = 60.0,
                     retry_policy: Optional[RetryPolicy] = None,
                     middlewares: Optional[List[ClientMiddleware]] = None,
                     pricing_model: str | None = None) -> "CohereBedrockEmbeddingClient":
         settings = BedrockSettings()
-        return cls(model, config=config, model_pricing=model_pricing, pricing_model=pricing_model, max_retries=max_retries,
+        return cls(model, config=config, pricing_override=pricing_override, pricing_model=pricing_model, max_retries=max_retries,
                    initial_retry_delay=initial_retry_delay, max_retry_delay=max_retry_delay, retry_policy=retry_policy,
                    region=region or settings.region, profile=profile or settings.profile, middlewares=middlewares)
 
@@ -118,7 +118,12 @@ class CohereBedrockEmbeddingClient(
             return await client.invoke_model(modelId=self.model, body=json.dumps(request), contentType="application/json", accept="application/json")
 
     def _estimate_cost(self, tokens: int) -> float:
-        pricing = self.model_pricing or PricingRegistry.resolve_cohere_bedrock_embedding_pricing(self.pricing_model or self.model)
+        if self.pricing_override is not None:
+            pricing = self.pricing_override
+        else:
+            pricing = PricingRegistry.resolve_cohere_bedrock_embedding_pricing(
+                self.pricing_model if self.pricing_model is not None else self.model
+            )
         return pricing.calculate_cost(tokens, 0)
 
     def _to_config(self) -> CohereBedrockEmbeddingClientConfig: