model-library 0.1.8__py3-none-any.whl → 0.1.10__py3-none-any.whl

model_library/base/base.py

@@ -76,6 +76,7 @@ class LLMConfig(BaseModel):
     top_k: int | None = None
     reasoning: bool = False
     reasoning_effort: str | bool | None = None
+    compute_effort: str | None = None
     supports_images: bool = False
     supports_files: bool = False
     supports_videos: bool = False
@@ -155,6 +156,7 @@ class LLM(ABC):
 
         self.reasoning: bool = config.reasoning
         self.reasoning_effort: str | bool | None = config.reasoning_effort
+        self.compute_effort: str | None = config.compute_effort
 
         self.supports_files: bool = config.supports_files
         self.supports_videos: bool = config.supports_videos

model_library/config/alibaba_models.yaml

@@ -37,6 +37,11 @@ qwen-3-max-models:
     supports:
       tools: true
       images: false
+    # only applies for <32K input tokens
+    # TODO: add thresholds
+    costs_per_million_token:
+      input: 1.2
+      output: 6.0
 
   alibaba/qwen3-max-2026-01-23:
     label: Qwen 3 Max Thinking

model_library/config/all_models.json

@@ -122,7 +122,10 @@
             "ignored_for_cost": false
         },
         "provider_properties": {},
-        "costs_per_million_token": {},
+        "costs_per_million_token": {
+            "input": 1.2,
+            "output": 6.0
+        },
         "alternative_keys": [],
         "default_parameters": {
             "temperature": 0.7
@@ -259,7 +262,9 @@
                 "write_markup": 1.0
             }
         },
-        "alternative_keys": [],
+        "alternative_keys": [
+            "vercel/zai/glm-4.7"
+        ],
         "default_parameters": {
             "temperature": 1.0,
             "top_p": 1.0
@@ -269,6 +274,94 @@
         "full_key": "zai/glm-4.7",
         "slug": "zai_glm-4.7"
     },
+    "vercel/zai/glm-4.7": {
+        "company": "zAI",
+        "label": "GLM 4.7",
+        "description": "Latest model from ZAI",
+        "release_date": "2025-12-22",
+        "open_source": true,
+        "documentation_url": "https://docs.z.ai/",
+        "properties": {
+            "context_window": 200000,
+            "max_tokens": 128000,
+            "training_cutoff": null,
+            "reasoning_model": true
+        },
+        "supports": {
+            "images": false,
+            "files": false,
+            "temperature": true,
+            "tools": true
+        },
+        "metadata": {
+            "deprecated": false,
+            "available_for_everyone": true,
+            "available_as_evaluator": false,
+            "ignored_for_cost": false
+        },
+        "provider_properties": {},
+        "costs_per_million_token": {
+            "input": 0.6,
+            "output": 2.2,
+            "cache": {
+                "read": 0.11,
+                "read_discount": 1.0,
+                "write_markup": 1.0
+            }
+        },
+        "alternative_keys": [],
+        "default_parameters": {
+            "temperature": 1.0,
+            "top_p": 1.0
+        },
+        "provider_endpoint": "zai/glm-4.7",
+        "provider_name": "vercel",
+        "full_key": "vercel/zai/glm-4.7",
+        "slug": "vercel_zai_glm-4.7"
+    },
+    "xiaomi/mimo-v2-flash": {
+        "company": "Xiaomi",
+        "label": "MiMo V2 Flash",
+        "description": "MiMo V2 Flash is Xiaomi's Mixture-of-Experts (MoE) language model with 309B total parameters and 15B active parameters. Designed for high-speed reasoning and agentic workflows, it utilizes a novel hybrid attention architecture and Multi-Token Prediction (MTP) to achieve state-of-the-art performance while significantly reducing inference costs.",
+        "release_date": "2025-12-17",
+        "open_source": true,
+        "documentation_url": "https://platform.xiaomimimo.com/#/docs/",
+        "properties": {
+            "context_window": 256000,
+            "max_tokens": 64000,
+            "training_cutoff": "December 2024"
+        },
+        "supports": {
+            "images": false,
+            "files": false,
+            "temperature": true,
+            "tools": true
+        },
+        "metadata": {
+            "deprecated": false,
+            "available_for_everyone": true,
+            "available_as_evaluator": false,
+            "ignored_for_cost": false
+        },
+        "provider_properties": {},
+        "costs_per_million_token": {
+            "input": 0.1,
+            "output": 0.3,
+            "cache": {
+                "read": 0.01,
+                "write_markup": 1.0
+            }
+        },
+        "alternative_keys": [],
+        "default_parameters": {
+            "temperature": 0.3,
+            "top_p": 0.95
+        },
+        "provider_endpoint": "mimo-v2-flash",
+        "provider_name": "xiaomi",
+        "full_key": "xiaomi/mimo-v2-flash",
+        "slug": "xiaomi_mimo-v2-flash"
+    },
     "google/gemini-3-flash-preview": {
         "company": "Google",
         "label": "Gemini 3 Flash (12/25)",
@@ -856,7 +949,9 @@
             "available_as_evaluator": false,
             "ignored_for_cost": false
         },
-        "provider_properties": {},
+        "provider_properties": {
+            "supports_compute_effort": true
+        },
         "costs_per_million_token": {
             "input": 5.0,
             "output": 25.0,
@@ -906,7 +1001,9 @@
             "available_as_evaluator": false,
             "ignored_for_cost": false
         },
-        "provider_properties": {},
+        "provider_properties": {
+            "supports_compute_effort": true
+        },
         "costs_per_million_token": {
             "input": 5.0,
             "output": 25.0,
@@ -2210,7 +2307,10 @@
             "ignored_for_cost": false
         },
         "provider_properties": {},
-        "costs_per_million_token": {},
+        "costs_per_million_token": {
+            "input": 1.2,
+            "output": 6.0
+        },
         "alternative_keys": [],
         "default_parameters": {
             "temperature": 0.7
@@ -2244,7 +2344,10 @@
             "ignored_for_cost": false
         },
         "provider_properties": {},
-        "costs_per_million_token": {},
+        "costs_per_million_token": {
+            "input": 1.2,
+            "output": 6.0
+        },
         "alternative_keys": [],
         "default_parameters": {
             "temperature": 0.7
@@ -2611,7 +2714,10 @@
             "ignored_for_cost": false
         },
         "provider_properties": {},
-        "costs_per_million_token": {},
+        "costs_per_million_token": {
+            "input": 1.2,
+            "output": 6.0
+        },
         "alternative_keys": [],
         "default_parameters": {
             "temperature": 0.7
@@ -3920,7 +4026,7 @@
             "tools": true
         },
         "metadata": {
-            "deprecated": false,
+            "deprecated": true,
             "available_for_everyone": true,
             "available_as_evaluator": false,
             "ignored_for_cost": false
@@ -4550,7 +4656,7 @@
             "tools": true
         },
         "metadata": {
-            "deprecated": false,
+            "deprecated": true,
             "available_for_everyone": true,
             "available_as_evaluator": false,
             "ignored_for_cost": false

model_library/config/anthropic_models.yaml

@@ -22,7 +22,6 @@ base-config:
   documentation_url: https://docs.claude.com/en/docs/about-claude/models/overview
   default_parameters:
     temperature: 1
-
 claude-4-models:
   base-config:
     supports:
@@ -40,6 +39,8 @@ claude-4-models:
       cache:
         read: 0.5
         write: 6.25
+    provider_properties:
+      supports_compute_effort: true
     alternative_keys:
       - anthropic/claude-opus-4-5-20251101-thinking:
           properties:

model_library/config/mistral_models.yaml

@@ -38,6 +38,8 @@ magistral-models:
     costs_per_million_token:
       input: 2
       output: 5
+    metadata:
+      deprecated: true
 
   mistralai/magistral-medium-2509:
     label: Magistral Medium 1.2 (09/2025)

model_library/config/together_models.yaml

@@ -45,6 +45,8 @@ kimi-models:
     costs_per_million_token:
       input: 1.00
       output: 3.00
+    metadata:
+      deprecated: true
 
 # Meta Llama Models
 llama-4-models:

model_library/config/xiaomi_models.yaml

@@ -0,0 +1,43 @@
+base-config:
+  company: Xiaomi
+  open_source: true
+  documentation_url: https://platform.xiaomimimo.com/#/docs/
+  supports:
+    images: false
+    files: false
+    tools: true
+  metadata:
+    available_as_evaluator: false
+    available_for_everyone: true
+    ignored_for_cost: false
+  properties:
+    training_cutoff: December 2024
+
+xiaomi-models:
+  base-config:
+    properties:
+      context_window: 256000
+    supports:
+      temperature: true
+      top_p: true
+    default_parameters:
+      temperature: 0.3
+      top_p: 0.95
+
+  xiaomi/mimo-v2-flash:
+    label: MiMo V2 Flash
+    description:
+      MiMo V2 Flash is Xiaomi's Mixture-of-Experts (MoE) language model with
+      309B total parameters and 15B active parameters. Designed for high-speed
+      reasoning and agentic workflows, it utilizes a novel hybrid attention
+      architecture and Multi-Token Prediction (MTP) to achieve state-of-the-art
+      performance while significantly reducing inference costs.
+    release_date: 2025-12-17
+    properties:
+      context_window: 256000
+      max_tokens: 64000
+    costs_per_million_token:
+      input: 0.10
+      output: 0.30
+      cache:
+        read: 0.01

model_library/config/zai_models.yaml

@@ -51,6 +51,9 @@ zai-models:
         read: 0.11
     default_parameters:
       temperature: 1
+    alternative_keys:
+      - vercel/zai/glm-4.7
+
   zai/glm-4.5:
     label: GLM 4.5
     description: "z.AI old model"

model_library/providers/anthropic.py

@@ -23,6 +23,7 @@ from model_library.base import (
     InputItem,
     LLMBatchMixin,
     LLMConfig,
+    ProviderConfig,
     QueryResult,
     QueryResultCost,
     QueryResultMetadata,
@@ -48,6 +49,11 @@ from model_library.utils import (
 )
 
 
+class AnthropicConfig(ProviderConfig):
+    supports_compute_effort: bool = False
+    supports_auto_thinking: bool = False
+
+
 class AnthropicBatchMixin(LLMBatchMixin):
     """Batch processing support for Anthropic's Message Batches API."""
 
@@ -251,6 +257,8 @@ class AnthropicBatchMixin(LLMBatchMixin):
 
 @register_provider("anthropic")
 class AnthropicModel(LLM):
+    provider_config = AnthropicConfig()
+
     def _get_default_api_key(self) -> str:
         if self.delegate_config:
             return self.delegate_config.api_key.get_secret_value()
@@ -547,13 +555,22 @@ class AnthropicModel(LLM):
         body["max_tokens"] = self.max_tokens
 
         if self.reasoning:
-            budget_tokens = kwargs.pop(
-                "budget_tokens", get_default_budget_tokens(self.max_tokens)
-            )
-            body["thinking"] = {
-                "type": "enabled",
-                "budget_tokens": budget_tokens,
-            }
+            if self.provider_config.supports_auto_thinking:
+                body["thinking"] = {"type": "auto"}
+            else:
+                budget_tokens = kwargs.pop(
+                    "budget_tokens", get_default_budget_tokens(self.max_tokens)
+                )
+                body["thinking"] = {
+                    "type": "enabled",
+                    "budget_tokens": budget_tokens,
+                }
+
+        # effort controls compute allocation for text, tool calls, and thinking. Opus-4.5+
+        # use instead of reasoning_effort with auto_thinking
+        if self.provider_config.supports_compute_effort and self.compute_effort:
+            # default is "high"
+            body["output_config"] = {"effort": self.compute_effort}
 
         # Thinking models don't support temperature: https://docs.claude.com/en/docs/build-with-claude/extended-thinking#feature-compatibility
         if self.supports_temperature and not self.reasoning:
@@ -599,7 +616,17 @@ class AnthropicModel(LLM):
 
         stream_kwargs = {**body}
         if is_anthropic_endpoint:
-            betas = ["files-api-2025-04-14", "interleaved-thinking-2025-05-14"]
+            betas = ["files-api-2025-04-14"]
+            if self.provider_config.supports_auto_thinking:
+                betas.extend(
+                    [
+                        "auto-thinking-2026-01-12",
+                        "effort-2025-11-24",
+                        "max-effort-2026-01-24",
+                    ]
+                )
+            else:
+                betas.extend(["interleaved-thinking-2025-05-14"])
             if "sonnet-4-5" in self.model_name:
                 betas.append("context-1m-2025-08-07")
             stream_kwargs["betas"] = betas

model_library/providers/google/google.py

@@ -58,6 +58,7 @@ from model_library.exceptions import (
     ImmediateRetryException,
     InvalidStructuredOutputError,
     ModelNoOutputError,
+    RetryException,
 )
 from model_library.providers.google.batch import GoogleBatchMixin
 from model_library.register_models import register_provider
@@ -195,7 +196,7 @@ class GoogleModel(LLM):
                     # id check
                     new_input.append(
                         Content(
-                            role="function",
+                            role="user",
                             parts=[
                                 Part.from_function_response(
                                     name=item.tool_call.name,
@@ -361,16 +362,21 @@ class GoogleModel(LLM):
         contents: list[Content | None] = []
         finish_reason: FinishReason | None = None
 
+        chunks: list[GenerateContentResponse] = []
+
         async for chunk in stream:
+            chunks.append(chunk)
             candidates = chunk.candidates
             if not candidates:
                 continue
 
             content = candidates[0].content
 
+            meaningful_content = False
             if content and content.parts:
                 for part in content.parts:
                     if part.function_call:
+                        meaningful_content = True
                         if not part.function_call.name:
                             raise Exception(f"Invalid function call: {part}")
 
@@ -387,21 +393,34 @@ class GoogleModel(LLM):
                     if not part.text:
                         continue
                     if part.thought:
+                        meaningful_content = True
                         reasoning += part.text
                     else:
+                        meaningful_content = True
                         text += part.text
 
             if chunk.usage_metadata:
                 metadata = chunk.usage_metadata
-            if content:
+            if content and meaningful_content:
                 contents.append(content)
             if candidates[0].finish_reason:
                 finish_reason = candidates[0].finish_reason
 
         if finish_reason != FinishReason.STOP:
-            self.logger.error(f"Unexpected finish reason: {finish_reason}")
+            self.logger.error(
+                f"Unexpected finish reason: {finish_reason}, chunks: {chunks}"
+            )
+
+        if finish_reason == FinishReason.MALFORMED_FUNCTION_CALL:
+            # gemini handles malformed function calls server side
+            # we don't want to return the content that was supposed to have a tool call, without that tool call
+            # and since we don't get any info on the params, we throw an error
+
+            self.logger.error("The function call was malformed")
+            raise RetryException("The function call was malformed")
 
         if not text and not reasoning and not tool_calls:
+            self.logger.error(f"Empty response. Chunks: {chunks}")
             raise ModelNoOutputError("Model returned empty response")
 
         result = QueryResult(

model_library/providers/openai.py

@@ -611,6 +611,12 @@ class OpenAIModel(LLM):
                         if tool_call_chunk.id and (
                             not raw_tool_calls
                             or raw_tool_calls[-1].id != tool_call_chunk.id
+                            or (
+                                raw_tool_calls[-1].id == tool_call_chunk.id
+                                and self.provider == "deepseek"
+                                and func
+                                and func.name
+                            )  # TODO: remove hotfix once deepseek fixes their stuff
                         ):
                             raw_tool_calls.append(
                                 ChatCompletionMessageToolCall(
@@ -625,7 +631,7 @@ class OpenAIModel(LLM):
                                 )
                             )
                         # accumulate delta
-                        elif func:
+                        elif func and raw_tool_calls:
                             if func.name:
                                 raw_tool_calls[-1].function.name = func.name
                             if func.arguments:

model_library/providers/vercel.py

@@ -0,0 +1,34 @@
+from typing import Literal
+
+from pydantic import SecretStr
+
+from model_library import model_library_settings
+from model_library.base import (
+    DelegateConfig,
+    DelegateOnly,
+    LLMConfig,
+)
+from model_library.register_models import register_provider
+
+
+@register_provider("vercel")
+class VercelModel(DelegateOnly):
+    def __init__(
+        self,
+        model_name: str,
+        provider: Literal["vercel"] = "vercel",
+        *,
+        config: LLMConfig | None = None,
+    ):
+        super().__init__(model_name, provider, config=config)
+
+        # https://vercel.com/docs/ai-gateway/sdks-and-apis#quick-start
+        self.init_delegate(
+            config=config,
+            delegate_config=DelegateConfig(
+                base_url="https://ai-gateway.vercel.sh/v1",
+                api_key=SecretStr(model_library_settings.VERCEL_API_KEY),
+            ),
+            use_completions=True,
+            delegate_provider="openai",
+        )

model_library/providers/xiaomi.py

@@ -0,0 +1,34 @@
+from typing import Literal
+
+from pydantic import SecretStr
+
+from model_library import model_library_settings
+from model_library.base import (
+    DelegateConfig,
+    DelegateOnly,
+    LLMConfig,
+)
+from model_library.register_models import register_provider
+
+
+@register_provider("xiaomi")
+class XiaomiModel(DelegateOnly):
+    def __init__(
+        self,
+        model_name: str,
+        provider: Literal["xiaomi"] = "xiaomi",
+        *,
+        config: LLMConfig | None = None,
+    ):
+        super().__init__(model_name, provider, config=config)
+
+        # https://platform.xiaomimimo.com/#/docs/quick-start/first-api-call
+        self.init_delegate(
+            config=config,
+            delegate_config=DelegateConfig(
+                base_url="https://api.xiaomimimo.com/v1",
+                api_key=SecretStr(model_library_settings.XIAOMI_API_KEY),
+            ),
+            use_completions=True,
+            delegate_provider="openai",
+        )

model_library/register_models.py

@@ -170,6 +170,7 @@ class DefaultParameters(BaseModel):
     top_p: float | None = None
     top_k: int | None = None
     reasoning_effort: str | bool | None = None
+    compute_effort: str | bool | None = None
 
 
 class RawModelConfig(BaseModel):
@@ -338,6 +339,10 @@ def _register_models() -> ModelRegistry:
                         copy.slug = key.replace("/", "_")
                         copy.full_key = key
                         copy.alternative_keys = []
+                        copy.provider_properties = ProviderProperties.model_validate(
+                            provider_properties
+                        )
+
                         registry[key] = copy
 
     return registry

model_library/registry_utils.py

@@ -1,9 +1,8 @@
+import tiktoken
 from functools import cache
 from pathlib import Path
 from typing import TypedDict
 
-import tiktoken
-
 from model_library.base import (
     LLM,
     LLMConfig,
@@ -235,6 +234,14 @@ def get_model_names(
     )
 
 
+"""
+everything below this comment is included for legacy support of caselaw/corpfin custom models.
+@orestes please remove this as part of the migration to a standard CorpFin harness.
+"""
+
+DEFAULT_CONTEXT_WINDOW = 128_000
+
+
 @cache
 def _get_tiktoken_encoder():
     """Get cached tiktoken encoder for consistent tokenization."""
@@ -257,7 +264,7 @@ def auto_trim_document(
         Trimmed document, or original document if trimming isn't needed
     """
 
-    max_tokens = get_max_document_tokens(model_name)
+    max_tokens = get_max_document_tokens(model_name) or DEFAULT_CONTEXT_WINDOW
 
     encoding = _get_tiktoken_encoder()
     tokens = encoding.encode(document)
@@ -284,5 +291,5 @@ def get_max_document_tokens(model_name: str, output_buffer: int = 10000) -> int:
     # Import here to avoid circular imports
     from model_library.utils import get_context_window_for_model
 
-    context_window = get_context_window_for_model(model_name)
+    context_window = get_context_window_for_model(model_name) or DEFAULT_CONTEXT_WINDOW
     return context_window - output_buffer

model_library/utils.py

@@ -1,6 +1,5 @@
 import logging
 from collections.abc import Mapping, Sequence
-
 import httpx
 from anthropic import AsyncAnthropic
 from openai import AsyncOpenAI
@@ -75,16 +74,15 @@ def create_anthropic_client_with_defaults(
     )
 
 
-def get_context_window_for_model(model_name: str, default: int = 128_000) -> int:
+def get_context_window_for_model(model_name: str) -> int | None:
     """
     Get the context window for a model by looking up its configuration from the registry.
 
     Args:
         model_name: The name of the model in the registry (e.g., "openai/gpt-4o-mini-2024-07-18" or "azure/gpt-4o-mini-2024-07-18")
-        default: Default context window to return if model not found or missing context_window
 
     Returns:
-        Context window size in tokens
+        Context window size in tokens (or `None` if not found)
     """
     # import here to avoid circular imports
     from model_library.register_models import get_model_registry
@@ -98,7 +96,6 @@ def get_context_window_for_model(model_name: str, default: int = 128_000) -> int
         return model_config.properties.context_window
     else:
         logger.warning(
-            f"Model {model_name} not found in registry or missing context_window, "
-            f"using default context length of {default}"
+            f"Model {model_name} not found in registry or missing context_window"
         )
-        return default
+        return None

{model_library-0.1.8.dist-info → model_library-0.1.10.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: model-library
-Version: 0.1.8
+Version: 0.1.10
 Summary: Model Library for vals.ai 
 Author-email: "Vals AI, Inc." <contact@vals.ai>
 License: MIT
@@ -21,7 +21,7 @@ Requires-Dist: mistralai<2.0,>=1.9.10
 Requires-Dist: xai-sdk<2.0,>=1.0.0
 Requires-Dist: ai21<5.0,>=4.3.0
 Requires-Dist: boto3<2.0,>=1.38.27
-Requires-Dist: google-genai[aiohttp]>=1.51.0
+Requires-Dist: google-genai[aiohttp]>=1.62.0
 Requires-Dist: google-cloud-storage>=1.26.0
 Requires-Dist: pytest-xdist>=3.8.0
 Dynamic: license-file

{model_library-0.1.8.dist-info → model_library-0.1.10.dist-info}/RECORD

@@ -4,12 +4,12 @@ model_library/file_utils.py,sha256=FAZRRtDT8c4Rjfoj64Te3knEHggXAAfRRuS8WLCsSe8,3
 model_library/logging.py,sha256=rZrrVQlEmyZzvKx6nIOR8bKHl49wQIIW5c36Zqcigm4,888
 model_library/model_utils.py,sha256=l8oCltGeimMGtnne_3Q1EguVtzCj61UMsLsma-1czwg,753
 model_library/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-model_library/register_models.py,sha256=-CtFggD296wqO2nd6yC5GIWBoi0YD0kDyG8bwxBa1ec,13696
-model_library/registry_utils.py,sha256=Qcu5du88RlcHo_mpZqOumLDie019sKQ_qX55XcjYWNc,9530
+model_library/register_models.py,sha256=xSIQDwsIuZNbOPD-EudnvxHXkCu1xnAWyLek_laGntU,13902
+model_library/registry_utils.py,sha256=UmFLG0PBk9x52nX7kXF0OYOiYNJT_RyaEN-JEthKAWk,9804
 model_library/settings.py,sha256=6xD9RADiiLWDk3xafYeCV9_4ZwxA2vYzpul-VP3qQlQ,867
-model_library/utils.py,sha256=uBLM9vfh2YG0pYsNSYCgmOc1GbE7zW9ykhBSlYJ3q0c,3302
+model_library/utils.py,sha256=NcSyQW6173HjExaNDsEQxtNbXVuW3dj-2x03URGYzSc,3152
 model_library/base/__init__.py,sha256=TtxCXGUtkEqWZNMMofLPuC4orN7Ja2hemtbtHitt_UA,266
-model_library/base/base.py,sha256=mB1oanWnXNk_C5M70LcwXQ1EvPdjFpcENq_Me2VvQJk,20588
+model_library/base/base.py,sha256=yZkRz2DllCoXZnuJ9t-9oa17jr6_fTpPqRDF919TwyI,20690
 model_library/base/batch.py,sha256=-jd6L0ECc5pkj73zoX2ZYcv_9iQdqxEi1kEilwaXWSA,2895
 model_library/base/delegate_only.py,sha256=LYmci02i3ShQuF_3N0gJsUySVvFB__PEAgTQQryp0nA,4764
 model_library/base/input.py,sha256=JrnvBZ_xLcEmaMjnOfUS6GFV0QWtCGpJq0RQQL2YBG8,1934
@@ -17,10 +17,10 @@ model_library/base/output.py,sha256=z8x3-0a3seRBuNwRDWgzdXjKS5bSASuR01qoRTmDoWI,
 model_library/base/utils.py,sha256=AjVeC41vg8vbGGL0fQmSGtnnTuZ589Xl7ljv0N1Jzd0,3330
 model_library/config/README.md,sha256=i8_wHnlI6uHIqWN9fYBkDCglZM2p5ZMVD3SLlxiwUVk,4274
 model_library/config/ai21labs_models.yaml,sha256=ZWHhk1cep2GQIYHqkTS_0152mF3oZg2tSzMPmvfMRSI,2478
-model_library/config/alibaba_models.yaml,sha256=KTNt9YJrAmgVKiQp3yeloiqaHg7PyobSK_97_p5kjbE,1826
-model_library/config/all_models.json,sha256=5L_V2DQsJTESTfluxfqUo6FJOr55EaF8z5dSSCaevWU,537531
+model_library/config/alibaba_models.yaml,sha256=TkGIPVW0cUVfDeCr9Pz5bWt4qUutrUamhJ0ukUz_dPg,1958
+model_library/config/all_models.json,sha256=Nv8EMbF95Qln7fmNxn9iFSrh1FxGNoM1_dvdQxgRe4M,540907
 model_library/config/amazon_models.yaml,sha256=HgLmhpfedHCQtkPEviEJCBbAb-dNQPOnVtf4UnwrDds,7654
-model_library/config/anthropic_models.yaml,sha256=bTc_3Oqn4wCdq-dcWcEfmXrPVZjcR8-V6pTao7sGa_E,10475
+model_library/config/anthropic_models.yaml,sha256=P_ij3Y_scivwsPhwk5KcuaBb9DYsNVl5lDHh_R56F5I,10535
 model_library/config/cohere_models.yaml,sha256=ZfWrS1K45Hxd5nT_gpP5YGAovJcBIlLNIdaRyE3V-7o,5022
 model_library/config/deepseek_models.yaml,sha256=4CCrf-4UPBgFCrS6CQa3vzNiaYlD4B6dFJFK_kIYBWY,1156
 model_library/config/dummy_model.yaml,sha256=lImYJBtBVJk_jgnLbkuSyOshQphVlYCMkw-UiJIBYhY,877
@@ -29,17 +29,18 @@ model_library/config/google_models.yaml,sha256=Rg127nsBbHpk62X7WBq2ckdHo0bwYM0NV
 model_library/config/inception_models.yaml,sha256=YCqfQlkH_pTdHIKee5dP_aRFXw_fTIEQCpUvX2bwO0M,560
 model_library/config/kimi_models.yaml,sha256=336jcbMhcKDHFNxzRIYxFTV7v2W_93y2sY-IkiYtkuo,1340
 model_library/config/minimax_models.yaml,sha256=gWTuTcl1-zyCF6KRuU6DSre2Cw5gXC-TeKV2Qp4StnQ,1263
-model_library/config/mistral_models.yaml,sha256=mYKYSzJl24lUiA_erSkom7nCBxAoeJ57Mi3090q1ArM,5162
+model_library/config/mistral_models.yaml,sha256=p5mlzImOBVno_NFk6jsA6jRVUHkc458kSqXuhDn2hsE,5199
 model_library/config/openai_models.yaml,sha256=pKsYFF4TnuWnAJVby6Bw6JFJhr6rbkt75N4VIg40iO4,25061
 model_library/config/perplexity_models.yaml,sha256=WUDqhLvnB0kQhCCwPv19FYLHozet3m33Spdo6bGff3Q,2336
-model_library/config/together_models.yaml,sha256=BeqRJonYzPvWwoLfkyH0RMRKBYUrCSEQhg_25Nvx97M,23867
+model_library/config/together_models.yaml,sha256=qPtYWZ5kb373vuDCrHjXrfU8oWUP-RjqKvxVqhKMyR0,23904
 model_library/config/xai_models.yaml,sha256=2KRNNQy3kV-4xxSfhj7Uhp9TZF-S5qPlM8Ef-04zv8Y,7985
-model_library/config/zai_models.yaml,sha256=-96aPrxpwWcEQD7vNZjDhTMi8lWYWl56AHYt-mwEkgM,2418
+model_library/config/xiaomi_models.yaml,sha256=G1TnVJ6uar69NY2t5vH2L_Fq1AINoNVNngzgL04T908,1166
+model_library/config/zai_models.yaml,sha256=2i4IXdSpta8uVY0q_OhHIfvHZ4o0jI6nvn1AMh0cqkw,2468
 model_library/providers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 model_library/providers/ai21labs.py,sha256=xXqgk-bAC6WTZ7rvN9eHoE_IUukaU_qF3eLw-TX8gTY,6356
 model_library/providers/alibaba.py,sha256=0oMgs09BmRknuY5E66OL987xjtIrLHMh2jzSce8UiCY,3226
 model_library/providers/amazon.py,sha256=VvC0-DoZWXx-obMMkw9YLQOwwcMNYnCSIGrMGkun7tI,15589
-model_library/providers/anthropic.py,sha256=SQuGxg_XH4g1nPDkrejYdW5pDRArm__IBUixYKlfWI8,25022
+model_library/providers/anthropic.py,sha256=a5_rO2nBirlc79gFpLkaUlC3dfSnk9crNlebOi_xYyU,26046
 model_library/providers/azure.py,sha256=mEwoGz8ajwI6MW464spPoa36uFsPWaG_4WtwQZWkxuE,1784
 model_library/providers/cohere.py,sha256=9ZGNekBlvubVw8o-emxdIhre_00iq5saENCVaD74Jqc,944
 model_library/providers/deepseek.py,sha256=S7Ud80vKVJMCkovaLEtZEPr1jI_yfOq-R0Ln-jWwqdo,1040
@@ -48,23 +49,25 @@ model_library/providers/inception.py,sha256=_UHWd8b3_YbR5tHXOGwPkz5YHEBXVaYM8bP0
 model_library/providers/kimi.py,sha256=r1K0mbaNY_OUTERLttoItcjWMb9wTu7ghEIP0kXblG0,1362
 model_library/providers/minimax.py,sha256=VTLzELrya4q_q5pdkmOA61kALbMu6ZkNxMdkuIZSDY8,1361
 model_library/providers/mistral.py,sha256=SkIHBkeDZuLfNYkvcm7holk63zNIIGJ8KmoXjQi399o,10299
-model_library/providers/openai.py,sha256=tDWozNzs8jcbtjQWEoMzYQq0qZObqOb_iK5AD-MVT3k,37618
+model_library/providers/openai.py,sha256=H1QfyXegHQihPeOK1NBw0zE_wX6LfARVGMrsiClIXSE,37982
 model_library/providers/openrouter.py,sha256=FRKyeUyCSSQFCvoKvVbKzYJ_SL60jEU-Y6wWDyvevYc,962
 model_library/providers/perplexity.py,sha256=GT7nfu42ATbG7Eu_J8nrIwJMr_BpFpLOaQVOQG4gtRk,961
 model_library/providers/together.py,sha256=M1xAsbBd-41a3yHj2MDrysrIav9pp7eqGYUo16a4L8s,1911
 model_library/providers/vals.py,sha256=lQXaQ6494r0Ujv2B9epLfVNdWk1nkMOZt1LKgAYsXhc,10084
+model_library/providers/vercel.py,sha256=CZOr7ps0WRUQnoLDoD6ICu7X-G7-dHtHDQ_8ciSqiak,951
 model_library/providers/xai.py,sha256=WzAV1WnUrx9UenLGkP77JslKDFVz1T0g25PfCYzMOlw,10687
+model_library/providers/xiaomi.py,sha256=OJKz_dbolWt3lPkllw91Ntx1uoLQL_eM8VOEccN_ZDs,954
 model_library/providers/zai.py,sha256=C-0Q41vEhsb6PIbFIW4kXGSx1-mTuB-czB7Vq1_KBqk,1924
 model_library/providers/google/__init__.py,sha256=ypuLVL_QJEQ7C3S47FhC9y4wyawYOdGikAViJmACI0U,115
 model_library/providers/google/batch.py,sha256=kqe9ponDe9Tkrh_-4kWd5-47wYf4V_XSKXZnrFOzAHc,10187
-model_library/providers/google/google.py,sha256=9sEBUmLqKy1M62kwc_1ViaTDoW6j5rGA3z9kgCn9ATY,18418
+model_library/providers/google/google.py,sha256=IrcfygqDcsaHcQAabMVZ65sTgmbpVU8ZarOj3mvWwXI,19304
 model_library/retriers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 model_library/retriers/backoff.py,sha256=vmpzLje51yfvSgAEST_yNrL87_J_0pmI-jwnRek5Ie0,2296
 model_library/retriers/base.py,sha256=zmiGZq4yvJx7aE0FD8rqZuJxZavhKTXLjt58z5TKJNw,6832
 model_library/retriers/token.py,sha256=ypHzuJjEORsv931sjEI1lejRO8Ju_ljSrCGV4pgaa34,16794
 model_library/retriers/utils.py,sha256=fhtQ64aT8mZcwZ8aTXnLRU4HVIf8JQ3CfmArvxZPWvQ,279
-model_library-0.1.8.dist-info/licenses/LICENSE,sha256=x6mf4o7U_wHaaqcfxoU-0R6uYJLbqL_TNuoULP3asaA,1070
-model_library-0.1.8.dist-info/METADATA,sha256=PlDQKGnuuNlRnlD5MiEUmc5egBmkFSjAGpJbup6sSAs,7024
-model_library-0.1.8.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-model_library-0.1.8.dist-info/top_level.txt,sha256=HtQYxA_7RP8UT35I6VcUw20L6edI0Zf2t5Ys1uDGVjs,14
-model_library-0.1.8.dist-info/RECORD,,
+model_library-0.1.10.dist-info/licenses/LICENSE,sha256=x6mf4o7U_wHaaqcfxoU-0R6uYJLbqL_TNuoULP3asaA,1070
+model_library-0.1.10.dist-info/METADATA,sha256=Z9yL5DixUa51LvxWseNBrAzv74zh0Is4HGOpHgI3Xs0,7025
+model_library-0.1.10.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+model_library-0.1.10.dist-info/top_level.txt,sha256=HtQYxA_7RP8UT35I6VcUw20L6edI0Zf2t5Ys1uDGVjs,14
+model_library-0.1.10.dist-info/RECORD,,