model-library 0.1.1__py3-none-any.whl → 0.1.3__py3-none-any.whl

model_library/config/anthropic_models.yaml

@@ -23,6 +23,32 @@ base-config:
     temperature: 1
 
 claude-4-models:
+  base-config:
+    class_properties:
+      supports_batch_requests: true
+
+  anthropic/claude-opus-4-5-20251101:
+    label: Claude Opus 4.5 (Nonthinking)
+    release_date: 2025-11-24
+    properties:
+      context_window: 200_000
+      max_token_output: 64_000
+      extending_thinking: 64_000
+    class_properties:
+      available_for_everyone: false
+    default_parameters:
+      max_output_tokens: 64_000
+    costs_per_million_token:
+      input: 15.0
+      output: 75.0
+      cache:
+        read: 1.5
+        write: 18.75
+    alternative_keys:
+      - anthropic/claude-opus-4-5-20251101-thinking:
+          properties:
+            reasoning_model: true
+
   anthropic/claude-opus-4-1-20250805:
     label: Claude Opus 4.1 (Nonthinking)
     description: Advanced model for specialized complex
@@ -193,11 +219,12 @@ claude-3-5-models:
     alternative_keys:
       - anthropic/claude-3-5-sonnet-latest
       - anthropic/claude-3.5-sonnet-latest
-  
+
   anthropic/claude-3-5-sonnet-20240620:
     label: Claude 3.5 Sonnet
     release_date: 2024-06-20
-    description: Claude Sonnet 3.5 (June 2024) variant for code and content generation,
+    description:
+      Claude Sonnet 3.5 (June 2024) variant for code and content generation,
       multilingual and vision-capable, deprecated.
     class_properties:
       deprecated: true
@@ -335,7 +362,7 @@ claude-2-models:
     costs_per_million_token:
       input: 8.0
       output: 24.0
-  
+
   anthropic/claude-1.3:
     label: Claude 1.3
     release_date: null

model_library/config/deepseek_models.yaml

@@ -10,7 +10,7 @@ base-config:
     ignored_for_cost: false
   properties:
     reasoning_model: false
-
+    
 deepseek-v3p2-exp-models:
   base-config:
     class_properties:
@@ -23,6 +23,8 @@ deepseek-v3p2-exp-models:
     costs_per_million_token:
       input: 0.28
       output: 0.42
+      cache:
+        read_discount: 0.1
 
   deepseek/deepseek-chat:
     label: DeepSeek V3.2-Exp (Nonthinking)

model_library/config/google_models.yaml

@@ -31,6 +31,55 @@ gemma-models:
       input: 0.00
       output: 0.00
 
+
+gemini-3-models:
+  base-config:
+    properties:
+      context_window: 1_048_576
+      max_token_output: 8_192
+      training_cutoff: "2025-01"
+    class_properties:
+      supports_images: true
+      supports_files: true
+      supports_videos: true
+      supports_tools: true
+      supports_batch_requests: true
+      supports_temperature: true
+    costs_per_million_token:
+      cache:
+        read_discount: 0.1
+    default_parameters:
+      temperature: 1
+      reasoning_effort: "high"
+
+  google/gemini-3-pro-preview:
+    label: Gemini 3 Pro (11/25)
+    description: Gemini 3 Pro, Google's most powerful model.
+    release_date: 2025-11-18
+    properties:
+      context_window: 1048576
+      max_token_output: 65536
+      training_cutoff: "2025-01"
+      reasoning_model: true
+    class_properties:
+      supports_images: true
+      supports_files: true
+      supports_videos: true
+      supports_tools: true
+      supports_batch_requests: true
+      supports_temperature: true
+    costs_per_million_token:
+      input: 2
+      output: 12
+      cache:
+        read: 0.20
+      context:
+        threshold: 200_000
+        input: 2.5
+        output: 15.0
+        cache:
+          read: 0.40
+
 gemini-2.5-models:
   base-config:
     properties:

model_library/config/openai_models.yaml

@@ -31,6 +31,31 @@ gpt-5-models:
       supports_temperature: false
       supports_files: true
       supports_tools: true
+  openai/gpt-5.1-codex:
+    label: GPT 5.1 Codex
+    documentation_url: https://platform.openai.com/docs/models/gpt-5.1-codex
+    description: OpenAI's latest coding model
+    release_date: 2025-11-13
+    costs_per_million_token:
+      input: 1.25
+      output: 10.0
+      cache:
+        read: 0.125
+    default_parameters:
+      max_output_tokens: 128_000
+
+  openai/gpt-5.1-codex-mini:
+    label: GPT 5.1 Codex Mini
+    documentation_url: https://platform.openai.com/docs/models/gpt-5.1-codex-mini
+    description: OpenAI's miniature coding model
+    release_date: 2025-11-13
+    costs_per_million_token:
+      input: 0.25
+      output: 2.00
+      cache:
+        read: 0.025
+    default_parameters:
+      max_output_tokens: 128_000
 
   openai/gpt-5-codex:
     label: GPT 5 Codex
@@ -48,7 +73,24 @@ gpt-5-models:
       available_as_evaluator: true
       supports_images: true
     default_parameters:
-      temperature: 1
+      max_output_tokens: 128_000
+
+  openai/gpt-5.1-2025-11-13:
+    label: GPT 5.1
+    documentation_url: https://platform.openai.com/docs/models/gpt-5.1
+    description: GPT-5.1 is OpenAI's flagship model for coding and agentic tasks with configurable reasoning and non-reasoning effort.
+    release_date: 2025-11-13
+    costs_per_million_token:
+      input: 1.25
+      output: 10
+      cache:
+        read: 0.125
+    properties:
+      training_cutoff: "2024-09"
+    class_properties:
+      available_as_evaluator: true
+      supports_images: true
+    default_parameters:
       max_output_tokens: 128_000
 
   openai/gpt-5-2025-08-07:
@@ -67,7 +109,6 @@ gpt-5-models:
       available_as_evaluator: true
       supports_images: true
     default_parameters:
-      temperature: 1
       max_output_tokens: 128_000
     alternative_keys:
       - azure/gpt-5-2025-08-07
@@ -85,7 +126,6 @@ gpt-5-models:
     properties:
       training_cutoff: "2024-05"
     default_parameters:
-      temperature: 1
       max_output_tokens: 128_000
     class_properties:
       supports_images: true
@@ -105,7 +145,6 @@ gpt-5-models:
     properties:
       training_cutoff: "2024-05"
     default_parameters:
-      temperature: 1
       max_output_tokens: 128_000
     class_properties:
       supports_images: true

model_library/config/together_models.yaml

@@ -28,6 +28,7 @@ kimi-models:
       supports_temperature: true
     default_parameters:
       temperature: 0.3
+      max_output_tokens: 16_384
 
   together/moonshotai/Kimi-K2-Instruct:
     label: Kimi K2 Instruct

model_library/config/xai_models.yaml

@@ -8,7 +8,7 @@ base-config:
     supports_files: false
     available_for_everyone: true
     ignored_for_cost: false
-    supports_tools: false
+    supports_tools: true
   properties:
     reasoning_model: false
 
@@ -33,7 +33,6 @@ xai-models:
       reasoning_model: true
     class_properties:
       supports_images: false
-      supports_tools: true
     costs_per_million_token:
       input: 0.20
       output: 1.50
@@ -81,6 +80,68 @@ xai-models:
       - grok/grok-4-fast
       - grok/grok-4-fast-reasoning-latest
 
+  grok/grok-4-1-fast-reasoning:
+    label: Grok 4.1 Fast (Reasoning)
+    description: ""
+    release_date: 2025-10-19
+    open_source: false
+    class_properties:
+      supports_images: true
+      available_as_evaluator: true
+      supports_metadata: true
+      supports_files: false
+      available_for_everyone: true
+      ignored_for_cost: false
+    properties:
+      context_window: 2_000_000
+      max_token_output: 2_000_000 # from openrouter
+      training_cutoff: null
+      reasoning_model: true
+    documentation_url: ""
+    costs_per_million_token:
+      input: 0.20
+      output: 0.5
+      cache:
+        read: 0.05
+      context:
+        threshold: 128_000
+        input: 0.4
+        output: 1.0
+    default_parameters:
+      temperature: 0.7
+      max_output_tokens: 128000
+
+  grok/grok-4-1-fast-non-reasoning:
+    label: Grok 4.1 Fast Non-Reasoning
+    description: ""
+    release_date: 2025-10-19
+    open_source: false
+    class_properties:
+      supports_images: true
+      available_as_evaluator: true
+      supports_metadata: true
+      supports_files: false
+      available_for_everyone: true
+      ignored_for_cost: false
+    properties:
+      context_window: 2_000_000
+      max_token_output: 2_000_000 # from openrouter
+      training_cutoff: null
+      reasoning_model: false
+    documentation_url: ""
+    costs_per_million_token:
+      input: 0.20
+      output: 0.5
+      cache:
+        read: 0.05
+      context:
+        threshold: 128_000
+        input: 0.4
+        output: 1.0
+    default_parameters:
+      temperature: 0.7
+      max_output_tokens: 128000
+
   grok/grok-4-fast-non-reasoning:
     label: Grok 4 Fast (Non-Reasoning)
     description: Cost-efficient model focused on speed and efficiency for straightforward tasks like summarization or classification without deep logical processing. Unified architecture with reasoning variant, steered via system prompts.
@@ -121,7 +182,6 @@ xai-models:
     class_properties:
       supports_images: true
       available_for_everyone: false
-      supports_tools: true
     properties:
       context_window: 256_000
       max_token_output: 128_000

model_library/exceptions.py

@@ -5,6 +5,7 @@ from typing import Any, Callable
 
 import backoff
 from ai21 import TooManyRequestsError as AI21RateLimitError
+from anthropic import InternalServerError
 from anthropic import RateLimitError as AnthropicRateLimitError
 from backoff._typing import Details
 from httpcore import ReadError as HTTPCoreReadError
@@ -166,6 +167,7 @@ RETRIABLE_EXCEPTIONS = [
     OpenAIUnprocessableEntityError,
     OpenAIAPIConnectionError,
     AnthropicRateLimitError,
+    InternalServerError,
     AI21RateLimitError,
     RemoteProtocolError,  # httpx connection closing when running models from sdk
     HTTPXReadError,
@@ -189,6 +191,9 @@ RETRIABLE_EXCEPTION_CODES = [
     "internal_error",
     "server_error",
     "overloaded",
+    "throttling",  # AWS throttling errors
+    "throttlingexception",  # AWS throttling errors
+    "internal server error",
 ]
 
 
@@ -237,8 +242,9 @@ def retry_llm_call(
     logger: logging.Logger,
     max_tries: int = RETRY_MAX_TRIES,
     max_time: float | None = None,
-    backoff_callback: Callable[[int, Exception | None, float, float], None]
-    | None = None,
+    backoff_callback: (
+        Callable[[int, Exception | None, float, float], None] | None
+    ) = None,
 ):
     def on_backoff(details: Details):
         exception = details.get("exception")

model_library/file_utils.py

@@ -56,7 +56,7 @@ def concat_images(
         new_width = int(combined_image.width * scale_factor)
         new_height = int(combined_image.height * scale_factor)
 
-        combined_image = combined_image.resize(
+        combined_image = combined_image.resize(  # type: ignore
             (new_width, new_height), Image.Resampling.LANCZOS
         )
 

model_library/providers/ai21labs.py

@@ -26,9 +26,11 @@ from model_library.exceptions import (
     MaxOutputTokensExceededError,
     ModelNoOutputError,
 )
+from model_library.register_models import register_provider
 from model_library.utils import default_httpx_client
 
 
+@register_provider("ai21labs")
 class AI21LabsModel(LLM):
     _client: AsyncAI21Client | None = None
 

model_library/providers/alibaba.py

@@ -1,29 +1,21 @@
-import io
-from typing import Any, Literal, Sequence
+from typing import Literal
 
 from typing_extensions import override
 
 from model_library import model_library_settings
 from model_library.base import (
-    LLM,
-    FileInput,
-    FileWithId,
-    InputItem,
+    DelegateOnly,
     LLMConfig,
-    QueryResult,
     QueryResultCost,
     QueryResultMetadata,
-    ToolDefinition,
 )
 from model_library.providers.openai import OpenAIModel
+from model_library.register_models import register_provider
 from model_library.utils import create_openai_client_with_defaults
 
 
-class AlibabaModel(LLM):
-    @override
-    def get_client(self) -> None:
-        raise NotImplementedError("Not implemented")
-
+@register_provider("alibaba")
+class AlibabaModel(DelegateOnly):
     def __init__(
         self,
         model_name: str,
@@ -32,23 +24,20 @@ class AlibabaModel(LLM):
         config: LLMConfig | None = None,
     ):
         super().__init__(model_name, provider, config=config)
-        self.native: bool = False
 
-        self.delegate: OpenAIModel | None = (
-            None
-            if self.native
-            else OpenAIModel(
-                model_name=model_name,
-                provider=provider,
-                config=config,
-                custom_client=create_openai_client_with_defaults(
-                    api_key=model_library_settings.DASHSCOPE_API_KEY,
-                    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
-                ),
-                use_completions=True,
-            )
+        # https://www.alibabacloud.com/help/en/model-studio/first-api-call-to-qwen
+        self.delegate = OpenAIModel(
+            model_name=self.model_name,
+            provider=self.provider,
+            config=config,
+            custom_client=create_openai_client_with_defaults(
+                api_key=model_library_settings.DASHSCOPE_API_KEY,
+                base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1",
+            ),
+            use_completions=True,
         )
 
+    @override
     async def _calculate_cost(
         self,
         metadata: QueryResultMetadata,
@@ -94,54 +83,3 @@ class AlibabaModel(LLM):
             else None,
             cache_write=None,
         )
-
-    @override
-    async def parse_input(
-        self,
-        input: Sequence[InputItem],
-        **kwargs: Any,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_image(
-        self,
-        image: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_file(
-        self,
-        file: FileInput,
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def parse_tools(
-        self,
-        tools: list[ToolDefinition],
-    ) -> Any:
-        raise NotImplementedError()
-
-    @override
-    async def upload_file(
-        self,
-        name: str,
-        mime: str,
-        bytes: io.BytesIO,
-        type: Literal["image", "file"] = "file",
-    ) -> FileWithId:
-        raise NotImplementedError()
-
-    @override
-    async def _query_impl(
-        self,
-        input: Sequence[InputItem],
-        *,
-        tools: list[ToolDefinition],
-        **kwargs: object,
-    ) -> QueryResult:
-        if self.delegate:
-            return await self.delegate_query(input, tools=tools, **kwargs)
-        raise NotImplementedError()

model_library/providers/amazon.py

@@ -31,8 +31,11 @@ from model_library.exceptions import (
     MaxOutputTokensExceededError,
 )
 from model_library.model_utils import get_default_budget_tokens
+from model_library.register_models import register_provider
 
 
+@register_provider("amazon")
+@register_provider("bedrock")
 class AmazonModel(LLM):
     _client: BaseClient | None = None