model-library 0.1.4__py3-none-any.whl → 0.1.5__py3-none-any.whl

model_library/base/output.py

@@ -9,9 +9,7 @@ from pydantic import BaseModel, Field, computed_field, field_validator
 from typing_extensions import override
 
 from model_library.base.input import InputItem, ToolCall
-from model_library.base.utils import (
-    sum_optional,
-)
+from model_library.base.utils import add_optional
 from model_library.utils import truncate_str
 
 
@@ -42,10 +40,14 @@ class QueryResultCost(BaseModel):
     reasoning: float | None = None
     cache_read: float | None = None
     cache_write: float | None = None
+    total_override: float | None = None
 
     @computed_field
     @property
     def total(self) -> float:
+        if self.total_override is not None:
+            return self.total_override
+
         return sum(
             filter(
                 None,
@@ -86,6 +88,16 @@ class QueryResultCost(BaseModel):
             )
         )
 
+    def __add__(self, other: "QueryResultCost") -> "QueryResultCost":
+        return QueryResultCost(
+            input=self.input + other.input,
+            output=self.output + other.output,
+            reasoning=add_optional(self.reasoning, other.reasoning),
+            cache_read=add_optional(self.cache_read, other.cache_read),
+            cache_write=add_optional(self.cache_write, other.cache_write),
+            total_override=add_optional(self.total_override, other.total_override),
+        )
+
     @override
     def __repr__(self):
         use_cents = self.total < 1
@@ -150,18 +162,20 @@ class QueryResultMetadata(BaseModel):
         return QueryResultMetadata(
             in_tokens=self.in_tokens + other.in_tokens,
             out_tokens=self.out_tokens + other.out_tokens,
-            reasoning_tokens=sum_optional(
-                self.reasoning_tokens, other.reasoning_tokens
+            reasoning_tokens=cast(
+                int | None, add_optional(self.reasoning_tokens, other.reasoning_tokens)
             ),
-            cache_read_tokens=sum_optional(
-                self.cache_read_tokens, other.cache_read_tokens
+            cache_read_tokens=cast(
+                int | None,
+                add_optional(self.cache_read_tokens, other.cache_read_tokens),
             ),
-            cache_write_tokens=sum_optional(
-                self.cache_write_tokens, other.cache_write_tokens
+            cache_write_tokens=cast(
+                int | None,
+                add_optional(self.cache_write_tokens, other.cache_write_tokens),
             ),
             duration_seconds=self.default_duration_seconds
             + other.default_duration_seconds,
-            cost=self.cost,
+            cost=cast(QueryResultCost | None, add_optional(self.cost, other.cost)),
         )
 
     @override

model_library/base/utils.py

@@ -1,4 +1,4 @@
-from typing import Sequence, cast
+from typing import Sequence, TypeVar, cast
 
 from model_library.base.input import (
     FileBase,
@@ -8,17 +8,39 @@ from model_library.base.input import (
     ToolResult,
 )
 from model_library.utils import truncate_str
+from pydantic import BaseModel
 
+T = TypeVar("T", bound=BaseModel)
 
-def sum_optional(a: int | None, b: int | None) -> int | None:
-    """Sum two optional integers, returning None if both are None.
+
+def add_optional(
+    a: int | float | T | None, b: int | float | T | None
+) -> int | float | T | None:
+    """Add two optional objects, returning None if both are None.
 
     Preserves None to indicate "unknown/not provided" when both inputs are None,
-    otherwise treats None as 0 for summation.
+    otherwise returns the non-None value or their sum.
     """
     if a is None and b is None:
         return None
-    return (a or 0) + (b or 0)
+
+    if a is None or b is None:
+        return a or b
+
+    if isinstance(a, (int, float)) and isinstance(b, (int, float)):
+        return a + b
+
+    # NOTE: Ensure that the subtypes are the same so we can use the __add__ method just from one
+    if type(a) is type(b):
+        add_method = getattr(a, "__add__", None)
+        if add_method is not None:
+            return add_method(b)
+    else:
+        raise ValueError(
+            f"Cannot add {type(a)} and {type(b)} because they are not the same subclass"
+        )
+
+    return None
 
 
 def get_pretty_input_types(input: Sequence["InputItem"], verbose: bool = False) -> str:

model_library/config/all_models.json

@@ -1,4 +1,54 @@
 {
+    "google/gemini-3-flash-preview": {
+        "company": "Google",
+        "label": "Gemini 3 Flash (12/25)",
+        "description": "Google's newest budget workhorse model",
+        "release_date": "2025-12-17",
+        "open_source": false,
+        "documentation_url": "https://ai.google.dev/gemini-api/docs/models",
+        "properties": {
+            "context_window": 1048576,
+            "max_tokens": 65536,
+            "training_cutoff": null,
+            "reasoning_model": true
+        },
+        "supports": {
+            "images": true,
+            "videos": true,
+            "files": true,
+            "batch": true,
+            "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": 0.5,
+            "output": 3.0,
+            "cache": {
+                "read_discount": 0.1,
+                "write_markup": 1.0
+            },
+            "batch": {
+                "input_discount": 0.5,
+                "output_discount": 0.5
+            }
+        },
+        "alternative_keys": [],
+        "default_parameters": {
+            "temperature": 1.0,
+            "reasoning_effort": "high"
+        },
+        "provider_endpoint": "gemini-3-flash-preview",
+        "provider_name": "google",
+        "full_key": "google/gemini-3-flash-preview",
+        "slug": "google_gemini-3-flash-preview"
+    },
     "openai/gpt-5.2-pro-2025-12-11": {
         "company": "OpenAI",
         "label": "GPT 5.2 Pro",
@@ -15428,7 +15478,7 @@
             "tools": false
         },
         "metadata": {
-            "deprecated": false,
+            "deprecated": true,
             "available_for_everyone": true,
             "available_as_evaluator": false,
             "ignored_for_cost": false

model_library/config/google_models.yaml

@@ -54,6 +54,21 @@ gemini-3-models:
       temperature: 1
       reasoning_effort: "high"
 
+  google/gemini-3-flash-preview:
+    label: Gemini 3 Flash (12/25)
+    description: Google's newest budget workhorse model
+    release_date: 2025-12-17
+    properties:
+      context_window: 1048576
+      max_tokens: 65536
+      reasoning_model: true
+    costs_per_million_token:
+      input: 0.50
+      output: 3.00
+    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.

model_library/config/perplexity_models.yaml

@@ -46,6 +46,8 @@ perplexity-models:
     label: Sonar Reasoning
     description: Reasoning-focused search model that exposes intermediate thinking for step-by-step answers.
     documentation_url: https://docs.perplexity.ai/models/models/sonar-reasoning
+    metadata:
+      deprecated: true
     properties:
       context_window: 128000
       reasoning_model: true

model_library/exceptions.py

@@ -183,8 +183,8 @@ RETRIABLE_EXCEPTION_CODES = [
     "server_error",
     "overloaded",
     "throttling",  # AWS throttling errors
-    "throttlingexception",  # AWS throttling errors
     "internal server error",
+    "InternalServerError",
 ]
 
 

model_library/registry_utils.py

@@ -1,9 +1,11 @@
 from functools import cache
 from pathlib import Path
+from typing import TypedDict
 
 import tiktoken
 
 from model_library.base import LLM, LLMConfig, ProviderConfig
+from model_library.base.output import QueryResultCost, QueryResultMetadata
 from model_library.register_models import (
     CostProperties,
     ModelConfig,
@@ -129,6 +131,64 @@ def get_model_cost(model_str: str) -> CostProperties | None:
     return model_config.costs_per_million_token
 
 
+class TokenDict(TypedDict, total=False):
+    """Token counts for cost calculation."""
+
+    in_tokens: int
+    out_tokens: int
+    reasoning_tokens: int | None
+    cache_read_tokens: int | None
+    cache_write_tokens: int | None
+
+
+async def recompute_cost(
+    model_str: str,
+    tokens: TokenDict,
+) -> QueryResultCost:
+    """
+    Recompute the cost for a model based on token information.
+
+    Uses the model provider's existing _calculate_cost method to ensure
+    provider-specific cost calculations are applied.
+
+    Args:
+        model_str: The model identifier (e.g., "openai/gpt-4o")
+        tokens: Dictionary containing token counts with keys:
+            - in_tokens (required): Number of input tokens
+            - out_tokens (required): Number of output tokens
+            - reasoning_tokens (optional): Number of reasoning tokens
+            - cache_read_tokens (optional): Number of cache read tokens
+            - cache_write_tokens (optional): Number of cache write tokens
+
+    Returns:
+        QueryResultCost with computed costs
+
+    Raises:
+        ValueError: If required token parameters are missing
+        Exception: If model not found in registry or costs not configured
+    """
+    if "in_tokens" not in tokens:
+        raise ValueError("Token dict must contain 'in_tokens'")
+    if "out_tokens" not in tokens:
+        raise ValueError("Token dict must contain 'out_tokens'")
+
+    model = get_registry_model(model_str)
+
+    metadata = QueryResultMetadata(
+        in_tokens=tokens["in_tokens"],
+        out_tokens=tokens["out_tokens"],
+        reasoning_tokens=tokens.get("reasoning_tokens"),
+        cache_read_tokens=tokens.get("cache_read_tokens"),
+        cache_write_tokens=tokens.get("cache_write_tokens"),
+    )
+
+    cost = await model._calculate_cost(metadata)  # type: ignore[arg-type]
+    if cost is None:
+        raise Exception(f"No cost information available for model {model_str}")
+
+    return cost
+
+
 @cache
 def get_provider_names() -> list[str]:
     """Return all provider names in the registry"""

{model_library-0.1.4.dist-info → model_library-0.1.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: model-library
-Version: 0.1.4
+Version: 0.1.5
 Summary: Model Library for vals.ai 
 Author-email: "Vals AI, Inc." <contact@vals.ai>
 License: MIT

{model_library-0.1.4.dist-info → model_library-0.1.5.dist-info}/RECORD

@@ -1,11 +1,11 @@
 model_library/__init__.py,sha256=AKc_15aklOf-LbcS9z1Xer_moRWNpG6Dh3kqvSQ0nOI,714
-model_library/exceptions.py,sha256=I9wquqj5hE640OfwVjUFtQUuu_potWAejLcOQCpDxIg,8705
+model_library/exceptions.py,sha256=ZHMr6lloXZz4V4Wy1UP8zc1CdUHx6-IS9_rOi6oN45s,8680
 model_library/file_utils.py,sha256=FAZRRtDT8c4Rjfoj64Te3knEHggXAAfRRuS8WLCsSe8,3682
 model_library/logging.py,sha256=McyaPHUk7RkB38-LrfnudrrU1B62ta8wAbbIBwLRmj0,853
 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=CY3Wd16AcWf7tYu_O2I2_kg_hdvQJFcvyQQA2OUu2SA,13646
-model_library/registry_utils.py,sha256=-ut95Aup5RYrZdv5Aih3bbYhe2vw9V0l4EFyH_1ONsQ,6797
+model_library/registry_utils.py,sha256=BVauHcP02Et2maLxowNBbdpGd32cnLz1_zSjDLVJjp0,8843
 model_library/settings.py,sha256=QyeUqzWBpexFi014L_mZkoXP49no3SAQNJRObATXrL8,873
 model_library/utils.py,sha256=T91ACGTc-KtksVyMFspt-vJtR5I-xcO3nVfH6SltmMU,3988
 model_library/base/__init__.py,sha256=TtxCXGUtkEqWZNMMofLPuC4orN7Ja2hemtbtHitt_UA,266
@@ -13,25 +13,25 @@ model_library/base/base.py,sha256=HXxImh2H-GIIiVGNqV7gRPi0HH1KJxB_4ckuKyEqAYo,14
 model_library/base/batch.py,sha256=-jd6L0ECc5pkj73zoX2ZYcv_9iQdqxEi1kEilwaXWSA,2895
 model_library/base/delegate_only.py,sha256=V2MzENtvBg0pySKncgE-mfCLBhhRZ0y4BntQwQsxbqU,2111
 model_library/base/input.py,sha256=Nhg8Ril1kFau1DnE8u102JC1l-vxNd-v9e3SjovR-Do,1876
-model_library/base/output.py,sha256=9pQZSOskkLDd_MAuDbYSimrbEcBL6x_3z6nLrPUnCOw,6701
-model_library/base/utils.py,sha256=KJZRVWr38Tik3yNJvTXnBy62ccilzzmSxHZFpQBJMPo,1330
+model_library/base/output.py,sha256=Ak6CJRYqtjYILsSWkfE70fSK3yvP7v_n5NYfysMaIL4,7464
+model_library/base/utils.py,sha256=YGQLPyQgCbfHNBxyTxCvpZNZ-ctEji258IdfMiXUJXs,1962
 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=-RLWOwh3ZaCQqjaZ-4Zw0BJNVE6JVHJ8Ggm9gQJZ6QI,2082
-model_library/config/all_models.json,sha256=HuTWNX-noeGfLNoWuzLVjhjXqkFGJX0CgBMt01Ejy3A,529312
+model_library/config/all_models.json,sha256=suyLLiU87NbUjWsnGeUpd94aIQ6UPyEURtPh48AW4-4,530914
 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/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
 model_library/config/fireworks_models.yaml,sha256=BMyQqjEpayNfSVGekzOFNIx7Ng3QOfPtldw5k2msqX0,6269
-model_library/config/google_models.yaml,sha256=n6yPRSVLyKGoJQW7L3UiVmb182zKiYhVLbmiUQDwXiY,16101
+model_library/config/google_models.yaml,sha256=Rg127nsBbHpk62X7WBq2ckdHo0bwYM0NVjF7T2h_1c0,16494
 model_library/config/inception_models.yaml,sha256=YCqfQlkH_pTdHIKee5dP_aRFXw_fTIEQCpUvX2bwO0M,560
 model_library/config/kimi_models.yaml,sha256=AAqse_BCE-lrHkJHIWJVqMtttnZQCa-5Qy5qiLUJjYs,755
 model_library/config/minimax_models.yaml,sha256=IttkpdBrp75J9WZQ0IRE4m4eSfd0LonfcA9OtrzJrMY,873
 model_library/config/mistral_models.yaml,sha256=mYKYSzJl24lUiA_erSkom7nCBxAoeJ57Mi3090q1ArM,5162
 model_library/config/openai_models.yaml,sha256=1lKsTQwsxMMJqXtEoYs3liy6NcaK4p8NN7b-GSFnl8k,25261
-model_library/config/perplexity_models.yaml,sha256=XEvs3fXrsSYjYNHLJuGSlTW7biHMaXpZaW4Q-aVn6wU,2299
+model_library/config/perplexity_models.yaml,sha256=WUDqhLvnB0kQhCCwPv19FYLHozet3m33Spdo6bGff3Q,2336
 model_library/config/together_models.yaml,sha256=BeqRJonYzPvWwoLfkyH0RMRKBYUrCSEQhg_25Nvx97M,23867
 model_library/config/xai_models.yaml,sha256=2KRNNQy3kV-4xxSfhj7Uhp9TZF-S5qPlM8Ef-04zv8Y,7985
 model_library/config/zai_models.yaml,sha256=Esa4P-zc5K1pejQTylKPe-uiH9AnvB_Zn7RB_sAZ5mU,1577
@@ -57,8 +57,8 @@ model_library/providers/zai.py,sha256=O_GM6KlJ0fM2wYoxO9xrCWfnpYH7IpoKEzjiD4jB8K
 model_library/providers/google/__init__.py,sha256=ypuLVL_QJEQ7C3S47FhC9y4wyawYOdGikAViJmACI0U,115
 model_library/providers/google/batch.py,sha256=4TE90Uo1adi54dVtGcGyUAxw11YExJq-Y4KmkQ-cyHA,9978
 model_library/providers/google/google.py,sha256=s9vky9r5SVNhBvMXcIr0_h0MlKLXwx_tQlZzs57xXYo,16507
-model_library-0.1.4.dist-info/licenses/LICENSE,sha256=x6mf4o7U_wHaaqcfxoU-0R6uYJLbqL_TNuoULP3asaA,1070
-model_library-0.1.4.dist-info/METADATA,sha256=4XPEbWSeOBYYoQ3ZYsdktZSnrDz2YbZixPIW7wTqJfw,6989
-model_library-0.1.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-model_library-0.1.4.dist-info/top_level.txt,sha256=HtQYxA_7RP8UT35I6VcUw20L6edI0Zf2t5Ys1uDGVjs,14
-model_library-0.1.4.dist-info/RECORD,,
+model_library-0.1.5.dist-info/licenses/LICENSE,sha256=x6mf4o7U_wHaaqcfxoU-0R6uYJLbqL_TNuoULP3asaA,1070
+model_library-0.1.5.dist-info/METADATA,sha256=HeLAgZOFNM7TBGJm2bfubjspsa388C0Va1hcWO-uu6I,6989
+model_library-0.1.5.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+model_library-0.1.5.dist-info/top_level.txt,sha256=HtQYxA_7RP8UT35I6VcUw20L6edI0Zf2t5Ys1uDGVjs,14
+model_library-0.1.5.dist-info/RECORD,,