llm-api-adapter 0.3.0__tar.gz → 0.3.2__tar.gz

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llm-api-adapter
-Version: 0.3.0
+Version: 0.3.2
 Summary: Lightweight, pluggable adapter for multiple LLM APIs (OpenAI, Anthropic, Google)
 Author: Sergey Inozemtsev
 License: MIT License
@@ -41,6 +41,7 @@ Dynamic: license-file
 # LLM API Adapter SDK for Python
 [![PyPI Downloads](https://static.pepy.tech/personalized-badge/llm-api-adapter?period=total&units=INTERNATIONAL_SYSTEM&left_color=BLACK&right_color=GREEN&left_text=downloads)](https://pepy.tech/projects/llm-api-adapter)
 [![Tests](https://github.com/Inozem/llm_api_adapter/actions/workflows/ci-main.yml/badge.svg)](https://github.com/Inozem/llm_api_adapter/actions/workflows/ci-main.yml)
+[![codecov](https://codecov.io/github/Inozem/llm_api_adapter/graph/badge.svg?token=T83ZPH1F7Z)](https://codecov.io/github/Inozem/llm_api_adapter)
 
 ## Overview
 
@@ -50,7 +51,7 @@ Currently, the project supports OpenAI, Anthropic, and Google with a consistent
 
 ### Version
 
-Current version: 0.3.0
+Current version: 0.3.2
 
 
 ## Features
@@ -183,9 +184,9 @@ The SDK provides a set of standardized errors for easier debugging and integrati
 
 The SDK allows you to easily switch between LLM providers and specify the model you want to use. Currently supported providers are OpenAI, Anthropic, and Google.
 
-- **OpenAI**: You can use models like `gpt-5.4`, `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-4.1`, `gpt-4.1-mini`, `gpt-4.1-nano`, `gpt-4o`, `gpt-4o-mini`.
-- **Anthropic**: Available models include `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-opus-4-5`, `claude-sonnet-4-5`, `claude-haiku-4-5`, `claude-opus-4-1`, `claude-opus-4-0`, `claude-sonnet-4-0`, `claude-3-haiku-20240307`.
-- **Google**: Models such as `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite-preview`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite` can be used.
+- **OpenAI**: You can use models like `gpt-5.5`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-4.1`, `gpt-4.1-mini`, `gpt-4.1-nano`, `gpt-4o`, `gpt-4o-mini`.
+- **Anthropic**: Available models include `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-opus-4-5`, `claude-sonnet-4-5`, `claude-haiku-4-5`, `claude-opus-4-1`.
+- **Google**: Models such as `gemini-3.5-flash`, `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash` can be used.
 
 Example:
 
@@ -487,6 +488,7 @@ first = adapter.chat(
     messages=messages,
     tools=tools,
     tool_choice="auto",
+    max_tokens=1000,
 )
 
 if first.tool_calls:
@@ -501,7 +503,11 @@ if first.tool_calls:
             )
         )
 
-    final = adapter.chat(messages=messages)
+    final = adapter.chat(
+        messages=messages,
+        previous_response=first,
+        max_tokens=1000
+    )
     print(final.content)
 ```
 

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/README.md

@@ -1,6 +1,7 @@
 # LLM API Adapter SDK for Python
 [![PyPI Downloads](https://static.pepy.tech/personalized-badge/llm-api-adapter?period=total&units=INTERNATIONAL_SYSTEM&left_color=BLACK&right_color=GREEN&left_text=downloads)](https://pepy.tech/projects/llm-api-adapter)
 [![Tests](https://github.com/Inozem/llm_api_adapter/actions/workflows/ci-main.yml/badge.svg)](https://github.com/Inozem/llm_api_adapter/actions/workflows/ci-main.yml)
+[![codecov](https://codecov.io/github/Inozem/llm_api_adapter/graph/badge.svg?token=T83ZPH1F7Z)](https://codecov.io/github/Inozem/llm_api_adapter)
 
 ## Overview
 
@@ -10,7 +11,7 @@ Currently, the project supports OpenAI, Anthropic, and Google with a consistent
 
 ### Version
 
-Current version: 0.3.0
+Current version: 0.3.2
 
 
 ## Features
@@ -143,9 +144,9 @@ The SDK provides a set of standardized errors for easier debugging and integrati
 
 The SDK allows you to easily switch between LLM providers and specify the model you want to use. Currently supported providers are OpenAI, Anthropic, and Google.
 
-- **OpenAI**: You can use models like `gpt-5.4`, `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-4.1`, `gpt-4.1-mini`, `gpt-4.1-nano`, `gpt-4o`, `gpt-4o-mini`.
-- **Anthropic**: Available models include `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-opus-4-5`, `claude-sonnet-4-5`, `claude-haiku-4-5`, `claude-opus-4-1`, `claude-opus-4-0`, `claude-sonnet-4-0`, `claude-3-haiku-20240307`.
-- **Google**: Models such as `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite-preview`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite` can be used.
+- **OpenAI**: You can use models like `gpt-5.5`, `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-4.1`, `gpt-4.1-mini`, `gpt-4.1-nano`, `gpt-4o`, `gpt-4o-mini`.
+- **Anthropic**: Available models include `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-opus-4-5`, `claude-sonnet-4-5`, `claude-haiku-4-5`, `claude-opus-4-1`.
+- **Google**: Models such as `gemini-3.5-flash`, `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash` can be used.
 
 Example:
 
@@ -447,6 +448,7 @@ first = adapter.chat(
     messages=messages,
     tools=tools,
     tool_choice="auto",
+    max_tokens=1000,
 )
 
 if first.tool_calls:
@@ -461,7 +463,11 @@ if first.tool_calls:
             )
         )
 
-    final = adapter.chat(messages=messages)
+    final = adapter.chat(
+        messages=messages,
+        previous_response=first,
+        max_tokens=1000
+    )
     print(final.content)
 ```
 

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "llm-api-adapter"
-version = "0.3.0"
+version = "0.3.2"
 description = "Lightweight, pluggable adapter for multiple LLM APIs (OpenAI, Anthropic, Google)"
 readme = "README.md"
 requires-python = ">=3.9"

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/adapters/anthropic_adapter.py

@@ -32,19 +32,19 @@ class AnthropicAdapter(LLMAdapterBase):
         tools: Optional[List[ToolSpec]] = None,
         tool_choice: Optional[str | dict] = None,
         parallel_tool_calls: Optional[bool] = None,
-        previous_response_id: Optional[str] = None,
+        previous_response: Optional[ChatResponse] = None,
     ) -> ChatResponse:
-        temperature = self._validate_parameter(
-            name="temperature", value=temperature, min_value=0, max_value=2
-        )
-        top_p = self._validate_parameter(name="top_p", value=top_p, min_value=0, max_value=1)
+        temperature = self._validate_parameter("temperature", temperature, 0, 2)
+        top_p = self._validate_parameter("top_p", top_p, 0, 1)
         try:
             self._validate_tools(tools)
             validated_tools = tools
-            normalized_tool_choice = self._normalize_tool_choice(tool_choice, validated_tools)
+            normalized_tool_choice = self._normalize_tool_choice(
+                tool_choice,
+                validated_tools,
+            )
             normalized_messages = self._normalize_messages(messages)
             system_prompt, transformed_messages = normalized_messages.to_anthropic()
-            client = ClaudeSyncClient(api_key=self.api_key)
             params: Dict[str, Any] = {
                 "model": self.model,
                 "messages": transformed_messages,
@@ -55,23 +55,36 @@ class AnthropicAdapter(LLMAdapterBase):
                 "timeout_s": timeout_s,
             }
             if validated_tools:
-                params["tools"] = [self._to_anthropic_tool(t) for t in validated_tools]
+                params["tools"] = [
+                    self._to_anthropic_tool(tool)
+                    for tool in validated_tools
+                ]
             if normalized_tool_choice is not None:
-                params["tool_choice"] = self._to_anthropic_tool_choice(normalized_tool_choice)
+                params["tool_choice"] = self._to_anthropic_tool_choice(
+                    normalized_tool_choice
+                )
             if parallel_tool_calls is False:
                 params["disable_parallel_tool_use"] = True
             elif parallel_tool_calls is True:
                 params["disable_parallel_tool_use"] = False
             if reasoning_level:
-                normalized_reasoning_level = self._normalize_reasoning_level(reasoning_level)
+                normalized_reasoning_level = self._normalize_reasoning_level(
+                    reasoning_level
+                )
                 if normalized_reasoning_level:
                     self.validate_reasoning_and_tokens(
                         max_tokens=max_tokens,
                         reasoning_level=reasoning_level,
                         normalized_reasoning_level=normalized_reasoning_level,
                     )
-                    params["thinking"] = {"type": "enabled", "budget_tokens": normalized_reasoning_level}
+                    params["budget_tokens"] = normalized_reasoning_level
+                if self.is_reasoning:
+                    effort = self._reasoning_level_to_effort(reasoning_level)
+                    if effort:
+                        params["effort"] = effort
             params = {k: v for k, v in params.items() if v is not None}
+            _ = previous_response
+            client = ClaudeSyncClient(api_key=self.api_key)
             response = client.chat_completion(**params)
             chat_response = ChatResponse.from_anthropic_response(response)
             if self.pricing:
@@ -88,12 +101,18 @@ class AnthropicAdapter(LLMAdapterBase):
             self.handle_error(error=e, error_message=error_message)
 
     def _to_anthropic_tool(self, tool: ToolSpec) -> Dict[str, Any]:
-        payload: Dict[str, Any] = {"name": tool.name, "input_schema": tool.json_schema}
+        payload: Dict[str, Any] = {
+            "name": tool.name,
+            "input_schema": tool.json_schema,
+        }
         if tool.description:
             payload["description"] = tool.description
         return payload
 
-    def _to_anthropic_tool_choice(self, tool_choice: Optional[str]) -> Optional[Dict[str, Any]]:
+    def _to_anthropic_tool_choice(
+        self,
+        tool_choice: Optional[str],
+    ) -> Optional[Dict[str, Any]]:
         if tool_choice is None:
             return None
         if tool_choice == "none":
@@ -102,42 +121,60 @@ class AnthropicAdapter(LLMAdapterBase):
             return {"type": tool_choice}
         return {"type": "tool", "name": tool_choice}
 
+    def _validate_not_bool(self, level) -> None:
+        if isinstance(level, bool):
+            raise ValueError("Invalid type for level: bool is not accepted")
+
+    def _validate_reasoning_str(self, level: str) -> None:
+        if level not in self.reasoning_levels:
+            raise ValueError(
+                f"Unknown reasoning level key: {level!r}. "
+                f"Valid keys: {list(self.reasoning_levels.keys())}"
+            )
+
+    def _resolve_reasoning_int(self, level: int) -> int:
+        if isinstance(level, int):
+            return level
+        raise ValueError(
+            "Invalid type for level: expected int or str, "
+            f"got {type(level).__name__!r}"
+        )
+
+    def _reasoning_level_to_effort(self, level: str | int) -> str | None:
+        self._validate_not_bool(level)
+        if isinstance(level, str):
+            self._validate_reasoning_str(level)
+            return None if level == "none" else level
+        numeric = self._resolve_reasoning_int(level)
+        for key, threshold in self.reasoning_levels.items():
+            if threshold > 0 and numeric <= threshold:
+                return key
+        return list(self.reasoning_levels)[-1]
+
     def _normalize_reasoning_level(self, level: str | int) -> int | None:
         minimum_level = 1024
-        normalized_level: int | None = None
-        if level is not None and not self.is_reasoning:
+        if not self.is_reasoning:
             warning_message = (
                 f"Model '{self.model}' does not support reasoning — reasoning disabled."
             )
             warnings.warn(warning_message, UserWarning)
             logger.info(warning_message)
             return None
-        if isinstance(level, bool):
-            raise ValueError("Invalid type for level: bool is not accepted")
+        self._validate_not_bool(level)
         if isinstance(level, str):
-            if level in self.reasoning_levels:
-                normalized_level = self.reasoning_levels[level]
-            else:
-                raise ValueError(
-                    f"Unknown reasoning level key: {level!r}. "
-                    f"Valid keys: {list(self.reasoning_levels.keys())}"
-                )
-        if isinstance(level, int):
-            normalized_level = level
-        if normalized_level is not None:
-            if normalized_level >= minimum_level:
-                return normalized_level
-            warning_message = (
-                f"Reasoning level '{level}' is below the minimum supported value {minimum_level}; "
-                f"using {minimum_level} instead."
-            )
-            warnings.warn(warning_message, UserWarning)
-            logger.info(warning_message)
-            return minimum_level
-        raise ValueError(
-            "Invalid type for level: expected int or str, "
-            f"got {type(level).__name__!r}"
+            self._validate_reasoning_str(level)
+            numeric = self.reasoning_levels[level]
+        else:
+            numeric = self._resolve_reasoning_int(level)
+        if numeric >= minimum_level:
+            return numeric
+        warning_message = (
+            f"Reasoning level '{level}' is below the minimum supported value "
+            f"{minimum_level}; using {minimum_level} instead."
         )
+        warnings.warn(warning_message, UserWarning)
+        logger.info(warning_message)
+        return minimum_level
 
     def validate_reasoning_and_tokens(
         self,

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/adapters/google_adapter.py

@@ -31,29 +31,39 @@ class GoogleAdapter(LLMAdapterBase):
         tools: Optional[List[ToolSpec]] = None,
         tool_choice: Optional[str | dict] = None,
         parallel_tool_calls: Optional[bool] = None,
-        previous_response_id: Optional[str] = None,
+        previous_response: Optional[ChatResponse] = None,
     ) -> ChatResponse:
         temperature = self._validate_parameter(
-            name="temperature", value=temperature, min_value=0, max_value=2
+            name="temperature",
+            value=temperature,
+            min_value=0,
+            max_value=2,
         )
         top_p = self._validate_parameter(
-            name="top_p", value=top_p, min_value=0, max_value=1
+            name="top_p",
+            value=top_p,
+            min_value=0,
+            max_value=1,
         )
         try:
             self._validate_tools(tools)
             validated_tools = tools
             normalized_tool_choice = self._normalize_tool_choice(
-                tool_choice, validated_tools
+                tool_choice,
+                validated_tools,
             )
             normalized_messages = self._normalize_messages(messages)
+            _ = previous_response
             system_prompt, transformed_messages = normalized_messages.to_google()
-            generation_config = {
+            generation_config: Dict[str, Any] = {
                 "maxOutputTokens": max_tokens,
                 "temperature": temperature,
                 "topP": top_p,
             }
             if reasoning_level:
-                normalized_reasoning_level = self._normalize_reasoning_level(reasoning_level)
+                normalized_reasoning_level = self._normalize_reasoning_level(
+                    reasoning_level
+                )
                 if normalized_reasoning_level is not None:
                     generation_config["thinkingConfig"] = {
                         "thinkingBudget": normalized_reasoning_level,
@@ -64,10 +74,17 @@ class GoogleAdapter(LLMAdapterBase):
                 "generationConfig": generation_config,
             }
             if system_prompt:
-                payload["system_instruction"] = {"parts": [{"text": system_prompt}]}
+                payload["system_instruction"] = {
+                    "parts": [{"text": system_prompt}]
+                }
             if validated_tools:
                 payload["tools"] = [
-                    {"functionDeclarations": [self._to_google_function_declaration(t) for t in validated_tools]}
+                    {
+                        "functionDeclarations": [
+                            self._to_google_function_declaration(tool)
+                            for tool in validated_tools
+                        ]
+                    }
                 ]
             tool_config = self._to_google_tool_config(normalized_tool_choice)
             if tool_config is not None:
@@ -94,34 +111,36 @@ class GoogleAdapter(LLMAdapterBase):
             self.handle_error(error=e, error_message=error_message)
 
     def _to_google_function_declaration(self, tool: ToolSpec) -> Dict[str, Any]:
-        decl: Dict[str, Any] = {
+        declaration: Dict[str, Any] = {
             "name": tool.name,
             "parametersJsonSchema": tool.json_schema,
         }
         if tool.description:
-            decl["description"] = tool.description
-        return decl
+            declaration["description"] = tool.description
+        return declaration
 
-    def _to_google_tool_config(self, tool_choice: Optional[str], ) -> Optional[Dict[str, Any]]:
+    def _to_google_tool_config(
+        self,
+        tool_choice: Optional[str],
+    ) -> Optional[Dict[str, Any]]:
         if tool_choice is None:
             return None
         if tool_choice == "none":
             mode = "NONE"
-            allowed = None
+            allowed_function_names = None
         elif tool_choice == "auto":
             mode = "AUTO"
-            allowed = None
+            allowed_function_names = None
         elif tool_choice == "any":
             mode = "ANY"
-            allowed = None
+            allowed_function_names = None
         else:
             mode = "ANY"
-            allowed = [tool_choice]
-
-        cfg: Dict[str, Any] = {"mode": mode}
-        if allowed:
-            cfg["allowedFunctionNames"] = allowed
-        return {"functionCallingConfig": cfg}
+            allowed_function_names = [tool_choice]
+        function_calling_config: Dict[str, Any] = {"mode": mode}
+        if allowed_function_names:
+            function_calling_config["allowedFunctionNames"] = allowed_function_names
+        return {"functionCallingConfig": function_calling_config}
 
     def _normalize_reasoning_level(self, level: str | int | None) -> int | None:
         minimum_level = 0

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/adapters/openai_adapter.py

@@ -30,7 +30,7 @@ class OpenAIAdapter(LLMAdapterBase):
         tools: Optional[List[ToolSpec]] = None,
         tool_choice: Any = None,
         parallel_tool_calls: Optional[bool] = None,
-        previous_response_id: Optional[str] = None,
+        previous_response: Optional[ChatResponse] = None,
     ) -> ChatResponse:
         temperature = self._validate_parameter(
             name="temperature",
@@ -51,7 +51,13 @@ class OpenAIAdapter(LLMAdapterBase):
             client = OpenAISyncClient(api_key=self.api_key)
             normalized_messages = self._normalize_messages(messages)
             use_responses_api = client._should_use_responses_api(self.model)
-            normalized_reasoning_level = self._normalize_reasoning_level(reasoning_level)
+            normalized_reasoning_level = self._normalize_reasoning_level(
+                reasoning_level
+            )
+
+            previous_response_id: Optional[str] = None
+            if previous_response is not None:
+                previous_response_id = previous_response.response_id
 
             if use_responses_api:
                 transformed_messages = normalized_messages.to_openai_responses_input()
@@ -111,21 +117,24 @@ class OpenAIAdapter(LLMAdapterBase):
             error_message = getattr(e, "text", None) or str(e)
             self.handle_error(error=e, error_message=error_message)
 
-    def _map_tools_to_openai(self, tools: Optional[List[ToolSpec]]) -> Optional[List[Dict[str, Any]]]:
+    def _map_tools_to_openai(
+        self,
+        tools: Optional[List[ToolSpec]],
+    ) -> Optional[List[Dict[str, Any]]]:
         if not tools:
             return None
         mapped: List[Dict[str, Any]] = []
-        for t in tools:
-            fn: Dict[str, Any] = {
-                "name": t.name,
-                "parameters": t.json_schema,
+        for tool in tools:
+            function_payload: Dict[str, Any] = {
+                "name": tool.name,
+                "parameters": tool.json_schema,
             }
-            if t.description:
-                fn["description"] = t.description
+            if tool.description:
+                function_payload["description"] = tool.description
             mapped.append(
                 {
                     "type": "function",
-                    "function": fn,
+                    "function": function_payload,
                 }
             )
         return mapped
@@ -167,7 +176,7 @@ class OpenAIAdapter(LLMAdapterBase):
             "Invalid type for level: expected int or str, "
             f"got {type(level).__name__!r}"
         )
-    
+
     def _map_tools_to_openai_responses(
         self,
         tools: Optional[List[ToolSpec]],
@@ -186,7 +195,7 @@ class OpenAIAdapter(LLMAdapterBase):
                 }
             )
         return result
-    
+
     def _map_tool_choice_to_openai_responses(self, tool_choice: Any) -> Any:
         if tool_choice is None:
             return None

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/llm_registry/llm_registry.json

@@ -1,14 +1,26 @@
 {
-  "schema_version": 5,
-  "effective_date": "2026-03-08",
+  "schema_version": 6,
+  "effective_date": "2026-06-22",
   "providers": {
     "openai": {
       "currency": "USD",
       "models": {
+        "gpt-5.5": {
+          "pricing": {"in_per_1m": 5.0, "out_per_1m": 30.0},
+          "is_reasoning": true
+        },
         "gpt-5.4": {
           "pricing": {"in_per_1m": 2.5, "out_per_1m": 15.0},
           "is_reasoning": true
         },
+        "gpt-5.4-mini": {
+          "pricing": {"in_per_1m": 0.75, "out_per_1m": 4.5},
+          "is_reasoning": true
+        },
+        "gpt-5.4-nano": {
+          "pricing": {"in_per_1m": 0.2, "out_per_1m": 1.25},
+          "is_reasoning": true
+        },
         "gpt-5.2": {
           "pricing": {"in_per_1m": 1.75, "out_per_1m": 14.0},
           "is_reasoning": true
@@ -39,6 +51,14 @@
     "anthropic": {
       "currency": "USD",
       "models": {
+        "claude-opus-4-8": {
+          "pricing": {"in_per_1m": 5.0, "out_per_1m": 25.0},
+          "is_reasoning": true
+        },
+        "claude-opus-4-7": {
+          "pricing": {"in_per_1m": 5.0, "out_per_1m": 25.0},
+          "is_reasoning": true
+        },
         "claude-opus-4-6": {
           "pricing": {"in_per_1m": 5.0, "out_per_1m": 25.0},
           "is_reasoning": true
@@ -62,26 +82,21 @@
         "claude-opus-4-1": {
           "pricing": {"in_per_1m": 15.0, "out_per_1m": 75.0},
           "is_reasoning": true
-        },
-        "claude-opus-4-0": {
-          "pricing": {"in_per_1m": 15.0, "out_per_1m": 75.0},
-          "is_reasoning": true
-        },
-        "claude-sonnet-4-0": {
-          "pricing": {"in_per_1m": 3.0, "out_per_1m": 15.0},
-          "is_reasoning": true
-        },
-        "claude-3-haiku-20240307": {"pricing": {"in_per_1m": 0.25, "out_per_1m": 1.25}}
+        }
       }
     },
     "google": {
       "currency": "USD",
       "models": {
+        "gemini-3.5-flash": {
+          "pricing": {"in_per_1m": 1.5, "out_per_1m": 9.0},
+          "is_reasoning": true
+        },
         "gemini-3.1-pro-preview": {
           "pricing": {"in_per_1m": 2.0, "out_per_1m": 12.0},
           "is_reasoning": true
         },
-        "gemini-3.1-flash-lite-preview": {
+        "gemini-3.1-flash-lite": {
           "pricing": {"in_per_1m": 0.25, "out_per_1m": 1.5},
           "is_reasoning": true
         },
@@ -96,10 +111,6 @@
         "gemini-2.5-flash": {
           "pricing": {"in_per_1m": 0.3, "out_per_1m": 2.5},
           "is_reasoning": true
-        },
-        "gemini-2.5-flash-lite": {
-          "pricing": {"in_per_1m": 0.1, "out_per_1m": 0.4},
-          "is_reasoning": true
         }
       }
     }

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/llms/anthropic/sync_client.py

@@ -14,6 +14,8 @@ from ...errors.llm_api_error import (
 
 logger = logging.getLogger(__name__)
 
+_ADAPTIVE_THINKING_MODELS = ("claude-opus-4-6", "claude-sonnet-4-6", "claude-opus-4-7", "claude-opus-4-8")
+
 @dataclass
 class ClaudeSyncClient:
     api_key: str
@@ -34,11 +36,20 @@ class ClaudeSyncClient:
         return response.json()
 
     def _prepare_chat_payload_for_model(self, model: str, kwargs: dict) -> dict:
-        if model.startswith(
-            ("claude-sonnet-4-5", "claude-opus-4-1", "claude-haiku-4-5", "claude-opus-4-5",
-             "claude-opus-4-6", "claude-sonnet-4-6")
-        ):
+        budget_tokens = kwargs.pop("budget_tokens", None)
+        effort = kwargs.pop("effort", None)
+        if model.startswith(_ADAPTIVE_THINKING_MODELS):
             kwargs.pop("top_p", None)
+            if effort:
+                kwargs["thinking"] = {"type": "adaptive"}
+                kwargs["output_config"] = {"effort": effort}
+        else:
+            if model.startswith(
+                ("claude-sonnet-4-5", "claude-opus-4-1", "claude-haiku-4-5", "claude-opus-4-5")
+            ):
+                kwargs.pop("top_p", None)
+            if budget_tokens:
+                kwargs["thinking"] = {"type": "enabled", "budget_tokens": budget_tokens}
         return {"model": model, **kwargs}
 
     def _send_request(self, url: str, payload: dict, timeout_s: float | None = None):

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/models/messages/chat_message.py

@@ -105,14 +105,17 @@ class AIMessage(Message):
             parts.append({"text": self.content})
         if self.tool_calls:
             for tc in self.tool_calls:
-                parts.append(
-                    {
-                        "functionCall": {
-                            "name": tc.name,
-                            "args": tc.arguments or {},
-                        }
+                part: Dict[str, Any] = {
+                    "functionCall": {
+                        "name": tc.name,
+                        "args": tc.arguments or {},
                     }
-                )
+                }
+                if tc.provider_data:
+                    thought_signature = tc.provider_data.get("thoughtSignature")
+                    if thought_signature is not None:
+                        part["thoughtSignature"] = thought_signature
+                parts.append(part)
         if not parts:
             parts = [{"text": ""}]
         return {"role": "model", "parts": parts}
@@ -262,10 +265,14 @@ class Messages:
         if not isinstance(arguments, dict):
             raise ValueError("assistant.tool_calls.arguments must be dict")
         call_id = tool_call_raw.get("call_id")
+        provider_data = tool_call_raw.get("provider_data")
+        if provider_data is not None and not isinstance(provider_data, dict):
+            raise ValueError("assistant.tool_calls.provider_data must be dict")
         return ToolCall(
             name=name,
             arguments=arguments,
             call_id=call_id,
+            provider_data=provider_data,
         )
 
     def _parse_legacy_openai_tool_call(self, tool_call_raw: Dict[str, Any]) -> ToolCall:

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/models/responses/chat_response.py

@@ -253,11 +253,18 @@ class ChatResponse:
                     raise InvalidToolArgumentsError(
                         detail=f"Google functionCall args must be dict for tool={name!r}"
                     )
+                thought_signature = part.get("thoughtSignature")
+                provider_data = None
+                if thought_signature is not None:
+                    provider_data = {
+                        "thoughtSignature": thought_signature,
+                    }
                 parsed_tool_calls.append(
                     ToolCall(
                         name=name,
                         arguments=args,
                         call_id=name,
+                        provider_data=provider_data,
                     )
                 )
         return cls(

{llm_api_adapter-0.3.0 → llm_api_adapter-0.3.2}/src/llm_api_adapter/models/tools/tool_call.py

@@ -14,3 +14,4 @@ class ToolCall:
     name: str
     arguments: Dict[str, Any]
     call_id: Optional[str] = None
+    provider_data: Optional[dict] = None