fix(vertex_ai): price web search on the Vertex Claude Sonnet 4.5 and Haiku 4.5 rows

Google bills Claude web search at $10 per 1,000 searches on these models, so token-priced deployments of them no longer bill searches at 0
This commit is contained in:
mateo-berri 2026-10-30 00:31:14 -07:00
parent 9f894486c9
commit 217fc1bc1b
3 changed files with 71 additions and 11 deletions

View file

@ -50998,6 +50998,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 5e-06,
"output_cost_per_token_batches": 2.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,
@ -51026,6 +51031,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 5e-06,
"output_cost_per_token_batches": 2.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,
@ -51772,6 +51782,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 1.5e-05,
"output_cost_per_token_batches": 7.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,
@ -51877,6 +51892,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 1.5e-05,
"output_cost_per_token_batches": 7.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,

View file

@ -50998,6 +50998,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 5e-06,
"output_cost_per_token_batches": 2.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,
@ -51026,6 +51031,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 5e-06,
"output_cost_per_token_batches": 2.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,
@ -51772,6 +51782,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 1.5e-05,
"output_cost_per_token_batches": 7.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,
@ -51877,6 +51892,11 @@
"max_input_tokens": 200000,
"max_output_tokens": 64000,
"mode": "chat",
"search_context_cost_per_query": {
"search_context_size_high": 0.01,
"search_context_size_low": 0.01,
"search_context_size_medium": 0.01
},
"output_cost_per_token": 1.5e-05,
"output_cost_per_token_batches": 7.5e-06,
"regional_endpoint_uplift_multiplier": 1.1,

View file

@ -1053,7 +1053,9 @@ def test_azure_ai_non_claude_web_search_keeps_generic_fallback() -> None:
assert get_cost_for_web_search_request(custom_llm_provider="azure_ai", usage=usage, model_info=model_info) is None
def _register_priced_vertex_deployment(model: str, search_cost_per_query: float | None) -> str:
def _register_priced_vertex_deployment(
model: str, search_cost_per_query: float | None, deployment_id: str | None = None
) -> str:
from litellm import Router
token_rates: Final = {"input_cost_per_token": 3e-6, "output_cost_per_token": 1.5e-5}
@ -1073,29 +1075,33 @@ def _register_priced_vertex_deployment(model: str, search_cost_per_query: float
{
"model_name": "priced-vertex-deployment",
"litellm_params": {"model": model, "vertex_project": "p", "vertex_location": "global"},
"model_info": {**token_rates, **search_rates},
"model_info": {**token_rates, **search_rates, **({} if deployment_id is None else {"id": deployment_id})},
}
]
)
return router.model_list[0]["model_info"]["id"]
def _vertex_claude_chat_response(web_search_requests: int) -> tuple[object, Usage | None]:
def _vertex_claude_chat_response(
web_search_requests: int, model: str = "claude-sonnet-5-5"
) -> tuple[object, Usage | None]:
usage: Final = Usage(
prompt_tokens=1840,
completion_tokens=412,
total_tokens=2252,
server_tool_use=ServerToolUse(web_search_requests=web_search_requests),
)
return ModelResponse(model="claude-sonnet-5-5", usage=usage), usage
return ModelResponse(model=model, usage=usage), usage
def _vertex_claude_messages_response(web_search_requests: int) -> tuple[object, Usage | None]:
def _vertex_claude_messages_response(
web_search_requests: int, model: str = "claude-sonnet-5-5"
) -> tuple[object, Usage | None]:
return {
"id": "msg_vertex_claude_web_search",
"type": "message",
"role": "assistant",
"model": "claude-sonnet-5-5",
"model": model,
"content": [{"type": "text", "text": "scripted answer"}],
"usage": {
"input_tokens": 1840,
@ -1128,19 +1134,33 @@ def test_vertex_claude_deployment_priced_web_search_billed_per_request(
@pytest.mark.parametrize("build_response", [_vertex_claude_chat_response, _vertex_claude_messages_response])
@pytest.mark.parametrize(
("deployment_model", "served_model"),
# Google bills Claude web search on Sonnet 4.5 and Haiku 4.5 too:
# https://cloud.google.com/model-garden/partner-models/pricing (read 2026-10-30)
[
("vertex_ai/claude-sonnet-5-5", "claude-sonnet-5-5"),
("vertex_ai/claude-sonnet-4-5@20250929", "claude-sonnet-4-5-20250929"),
("vertex_ai/claude-haiku-4-5@20251001", "claude-haiku-4-5-20251001"),
],
)
@pytest.mark.parametrize("deployment_id", [None, "vertex-claude-token-priced"], ids=["hashed_id", "claude_in_id"])
def test_vertex_claude_deployment_with_token_rates_only_bills_served_model_search_price(
local_model_cost_map: None,
build_response: Callable[[int], tuple[object, Usage | None]],
build_response: Callable[[int, str], tuple[object, Usage | None]],
deployment_model: str,
served_model: str,
deployment_id: str | None,
) -> None:
served_search_pricing: Final = (
litellm.get_model_info(model="vertex_ai/claude-sonnet-5-5").get("search_context_cost_per_query") or {}
litellm.get_model_info(model=deployment_model).get("search_context_cost_per_query") or {}
)
served_search_price: Final = served_search_pricing.get("search_context_size_medium") or 0.0
deployment_id: Final = _register_priced_vertex_deployment("vertex_ai/claude-sonnet-5-5", None)
response_object, usage = build_response(2)
registered_id: Final = _register_priced_vertex_deployment(deployment_model, None, deployment_id)
response_object, usage = build_response(2, served_model)
cost: Final = StandardBuiltInToolCostTracking.get_cost_for_built_in_tools(
model=deployment_id,
model=registered_id,
response_object=response_object,
usage=usage,
custom_llm_provider="vertex_ai",