diff --git a/tests/e2e/AGENTS.md b/tests/e2e/AGENTS.md
index cf1e41a59a6..dee7e88443f 100644
--- a/tests/e2e/AGENTS.md
+++ b/tests/e2e/AGENTS.md
@@ -75,7 +75,7 @@ class TestPromptCompression:
 
     def test_prompt_compression_accumulate_spend(self, key_id, user_id):
         for _ in range(10):
-            response = self.resources.proxy.post("gemini-2.5-flash", key_id, user_id)
+            response = self.resources.proxy.post("gemini-3.8-flash", key_id, user_id)
         compressed_value = ...
         assert response.cost == compressed_value  # the cost was actually reduced
 ```
diff --git a/tests/e2e/CONTRIBUTING.md b/tests/e2e/CONTRIBUTING.md
index b4379425854..53613313def 100644
--- a/tests/e2e/CONTRIBUTING.md
+++ b/tests/e2e/CONTRIBUTING.md
@@ -9,7 +9,7 @@ When contributing to this directory, please first discuss the change you wish to
 
 ## Setup
 
-The suites run against a live proxy, so bring one up first by running the litellm proxy locally. Point it at a config that prewires the example models the suites use (`gpt-5.5`, `claude-haiku-4-5`, `gemini-2.5-flash`, `openai-text-embedding-3-small`) with keys from your `.env`, and enables prompt storage, a redis cache, the fast budget rescheduler the quota suites rely on, and `router_settings.optional_pre_call_checks: ["prompt_caching"]`, which the router suite's prompt-cache affinity test reads back from `GET /router/settings` and fails without. If your test needs another model, a pricing override, or a guardrail declared up front, add it to that config and read it back in the test rather than hardcoding values
+The suites run against a live proxy, so bring one up first by running the litellm proxy locally. Point it at a config that prewires the example models the suites use (`gpt-5.5`, `claude-haiku-4-5`, `gemini-3.8-flash`, `openai-text-embedding-3-small`) with keys from your `.env`, and enables prompt storage, a redis cache, the fast budget rescheduler the quota suites rely on, and `router_settings.optional_pre_call_checks: ["prompt_caching"]`, which the router suite's prompt-cache affinity test reads back from `GET /router/settings` and fails without. If your test needs another model, a pricing override, or a guardrail declared up front, add it to that config and read it back in the test rather than hardcoding values
 
 ## Running the tests locally
 
@@ -206,7 +206,7 @@ class TestPromptCompression:
 
     def test_prompt_compression_accumulate_spend(self, key_id, user_id):
         for _ in range(10):
-            response = self.resources.proxy.post("gemini-2.5-flash", key_id, user_id)
+            response = self.resources.proxy.post("gemini-3.8-flash", key_id, user_id)
         compressed_value = ...
         assert response.cost == compressed_value  # the cost was actually reduced
 ```
diff --git a/tests/e2e/access_control/test_access_control_e2e.py b/tests/e2e/access_control/test_access_control_e2e.py
index 419cae78df2..b1965748197 100644
--- a/tests/e2e/access_control/test_access_control_e2e.py
+++ b/tests/e2e/access_control/test_access_control_e2e.py
@@ -30,7 +30,7 @@ from e2e_metadata import Domain, Mode, Provider, Route, Subject, meta
 
 pytestmark = pytest.mark.e2e
 
-ALLOWED_MODEL = "gemini-2.5-flash"
+ALLOWED_MODEL = "gemini-3.8-flash"
 DISALLOWED_MODEL = "gpt-5.5"
 VIRTUAL_KEY_BACKEND = "anthropic/claude-haiku-4-5-20251001"
 EMBEDDING_MODEL = "openai-text-embedding-3-small"
diff --git a/tests/e2e/access_control/test_project_all_team_models_e2e.py b/tests/e2e/access_control/test_project_all_team_models_e2e.py
index d70e3de554a..410578ece08 100644
--- a/tests/e2e/access_control/test_project_all_team_models_e2e.py
+++ b/tests/e2e/access_control/test_project_all_team_models_e2e.py
@@ -28,7 +28,7 @@ from models import ChatResponse, TeamInfoResponse
 
 pytestmark = pytest.mark.e2e
 
-TEAM_MODEL = "gemini-2.5-flash"
+TEAM_MODEL = "gemini-3.8-flash"
 OUTSIDE_MODEL = "gpt-5.5"
 
 
diff --git a/tests/e2e/batches/capabilities.py b/tests/e2e/batches/capabilities.py
index c03b481f060..90d80bc1bb6 100644
--- a/tests/e2e/batches/capabilities.py
+++ b/tests/e2e/batches/capabilities.py
@@ -23,7 +23,7 @@ def batch_model_name(base: str) -> str:
 
 OPENAI_BATCH_BACKEND: Final = "gpt-4o-mini"
 AZURE_BATCH_BACKEND: Final = "gpt-5.4-mini-batch"
-VERTEX_BATCH_BACKEND: Final = "gemini-2.5-flash"
+VERTEX_BATCH_BACKEND: Final = "gemini-3.8-flash"
 BEDROCK_BATCH_BACKEND: Final = "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0"
 
 
@@ -81,7 +81,7 @@ class Provider:
                 return LiteLLMParamsBody(
                     model=f"vertex_ai/{VERTEX_BATCH_BACKEND}",
                     vertex_project="os.environ/VERTEXAI_PROJECT",
-                    vertex_location="us-central1",
+                    vertex_location="global",
                     vertex_credentials="os.environ/VERTEXAI_CREDENTIALS",
                     gcs_bucket_name="os.environ/GCS_BUCKET_NAME",
                     bucket_name="os.environ/GCS_BUCKET_NAME",
diff --git a/tests/e2e/batches/test_batches_e2e.py b/tests/e2e/batches/test_batches_e2e.py
index b7a52da5902..8d76e5d9693 100644
--- a/tests/e2e/batches/test_batches_e2e.py
+++ b/tests/e2e/batches/test_batches_e2e.py
@@ -1286,7 +1286,7 @@ class TestBedrockBatchGovCloud:
         assert fetched.id == batch.id
 
 
-GEMINI_FILES_RAW_MODEL = "gemini-2.5-flash"
+GEMINI_FILES_RAW_MODEL = "gemini-3.8-flash"
 
 
 class TestGeminiFiles:
diff --git a/tests/e2e/gateway/stage_mirror_ci_config.yml b/tests/e2e/gateway/stage_mirror_ci_config.yml
index 02a131f60ae..983214e8e45 100644
--- a/tests/e2e/gateway/stage_mirror_ci_config.yml
+++ b/tests/e2e/gateway/stage_mirror_ci_config.yml
@@ -55,15 +55,15 @@ model_list:
     litellm_params:
       model: anthropic/claude-haiku-4-5
       api_key: os.environ/ANTHROPIC_API_KEY
-  - model_name: gemini-2.5-flash-vertex
+  - model_name: gemini-3.8-flash-vertex
     litellm_params:
-      model: vertex_ai/gemini-2.5-flash
+      model: vertex_ai/gemini-3.8-flash
       vertex_project: os.environ/VERTEXAI_PROJECT
-      vertex_location: us-central1
+      vertex_location: global
       vertex_credentials: os.environ/VERTEXAI_CREDENTIALS
-  - model_name: gemini-2.5-flash
+  - model_name: gemini-3.8-flash
     litellm_params:
-      model: gemini/gemini-2.5-flash
+      model: gemini/gemini-3.8-flash
       api_key: os.environ/GEMINI_API_KEY
   - model_name: e2e-alias-rl-target
     litellm_params:
diff --git a/tests/e2e/guardrails/guardrails_client.py b/tests/e2e/guardrails/guardrails_client.py
index 08124a03313..a06d86798f8 100644
--- a/tests/e2e/guardrails/guardrails_client.py
+++ b/tests/e2e/guardrails/guardrails_client.py
@@ -38,7 +38,7 @@ from models import (
 from proxy_client import ProxyClient
 from pydantic import BaseModel, Field
 
-GUARDRAIL_BACKEND: Final = "gemini/gemini-2.5-flash"
+GUARDRAIL_BACKEND: Final = "gemini/gemini-3.8-flash"
 
 GuardrailMode = Literal["pre_call", "post_call", "during_call", "logging_only"]
 PiiEntity = Literal["EMAIL_ADDRESS", "PHONE_NUMBER", "PERSON", "CREDIT_CARD", "US_SSN"]
diff --git a/tests/e2e/guardrails/test_bedrock_guardrail_e2e.py b/tests/e2e/guardrails/test_bedrock_guardrail_e2e.py
index 8f52eb56d5f..bd74a8a3b9d 100644
--- a/tests/e2e/guardrails/test_bedrock_guardrail_e2e.py
+++ b/tests/e2e/guardrails/test_bedrock_guardrail_e2e.py
@@ -48,7 +48,7 @@ def _without_assessments(value: JsonValue) -> JsonValue:
     return value
 
 
-MODEL = "gemini-2.5-flash"
+MODEL = "gemini-3.8-flash"
 # Matches the word/topic policy the guardrail this suite points at actually denies.
 # Content filters are not assumed: the guardrail resource carries no contentPolicy,
 # so a hate/violence prompt returns action=NONE and never blocks.
diff --git a/tests/e2e/guardrails/test_guardrail_dispatch_e2e.py b/tests/e2e/guardrails/test_guardrail_dispatch_e2e.py
index c8d295ee459..1ce41dd8ff7 100644
--- a/tests/e2e/guardrails/test_guardrail_dispatch_e2e.py
+++ b/tests/e2e/guardrails/test_guardrail_dispatch_e2e.py
@@ -16,7 +16,7 @@ from guardrails_client import GuardrailsClient
 
 pytestmark = pytest.mark.e2e
 
-MODEL = "gemini-2.5-flash"
+MODEL = "gemini-3.8-flash"
 
 
 @pytest.mark.skip(
diff --git a/tests/e2e/guardrails/test_key_guardrail_image_edit_e2e.py b/tests/e2e/guardrails/test_key_guardrail_image_edit_e2e.py
index 153d5fd0aa7..f26b0adeb54 100644
--- a/tests/e2e/guardrails/test_key_guardrail_image_edit_e2e.py
+++ b/tests/e2e/guardrails/test_key_guardrail_image_edit_e2e.py
@@ -13,7 +13,7 @@ from models import LiteLLMParamsBody
 
 pytestmark = pytest.mark.e2e
 
-CHAT_MODEL: Final = "gemini-2.5-flash"
+CHAT_MODEL: Final = "gemini-3.8-flash"
 IMAGE_BACKEND: Final = "openai/gpt-image-2.5-flare"
 SOURCE_PNG: Final = base64.b64decode(
     "iVBORw0KGgoAAAANSUhEUgAAAEAAAABACAIAAAAlC+aJAAAAS0lEQVR42u3PMQ0AAAwDoPo3"
diff --git a/tests/e2e/guardrails/test_key_guardrail_video_e2e.py b/tests/e2e/guardrails/test_key_guardrail_video_e2e.py
index 233cbfb8881..4a8affb5259 100644
--- a/tests/e2e/guardrails/test_key_guardrail_video_e2e.py
+++ b/tests/e2e/guardrails/test_key_guardrail_video_e2e.py
@@ -10,7 +10,7 @@ from models import LiteLLMParamsBody
 
 pytestmark = pytest.mark.e2e
 
-CHAT_MODEL = "gemini-2.5-flash"
+CHAT_MODEL = "gemini-3.8-flash"
 VIDEO_BACKEND = "vertex_ai/veo-3.1-fast-generate-001"
 
 
diff --git a/tests/e2e/guardrails/test_openai_moderation_category_matrix_e2e.py b/tests/e2e/guardrails/test_openai_moderation_category_matrix_e2e.py
index 10155cf96a5..df60e44135e 100644
--- a/tests/e2e/guardrails/test_openai_moderation_category_matrix_e2e.py
+++ b/tests/e2e/guardrails/test_openai_moderation_category_matrix_e2e.py
@@ -19,7 +19,7 @@ from models import AnthropicMessagesResponse, ChatResponse
 
 pytestmark = pytest.mark.e2e
 
-GEMINI_BACKEND: Final = "gemini/gemini-2.5-flash"
+GEMINI_BACKEND: Final = "gemini/gemini-3.8-flash"
 ANTHROPIC_BACKEND: Final = "anthropic/claude-haiku-4-5"
 OPENAI_BACKEND: Final = "openai/gpt-4o-mini"
 
diff --git a/tests/e2e/guardrails/test_presidio_masking_e2e.py b/tests/e2e/guardrails/test_presidio_masking_e2e.py
index 9fc2c7d3ea2..ddeff9bc6e2 100644
--- a/tests/e2e/guardrails/test_presidio_masking_e2e.py
+++ b/tests/e2e/guardrails/test_presidio_masking_e2e.py
@@ -53,7 +53,7 @@ from models import (
 
 pytestmark = pytest.mark.e2e
 
-MODEL = "gemini-2.5-flash"
+MODEL = "gemini-3.8-flash"
 
 # A guardrail created via POST /guardrails reaches the worker that served the
 # create immediately, but every other worker only picks it up on its next
@@ -268,7 +268,7 @@ class TestPresidioPreCallMasking:
         prompt = _pii_prompt(unique_marker(), email)
 
         _assert_eventually_masked(
-            lambda: client.chat(scoped_key, MODEL, prompt, guardrails=[name], max_tokens=128),
+            lambda: client.chat(scoped_key, MODEL, prompt, guardrails=[name], max_tokens=512),
             _first_content,
             email=email,
         )
@@ -296,7 +296,7 @@ class TestPresidioPreCallMasking:
         prompt = _pii_prompt(unique_marker(), email)
 
         _assert_eventually_masked(
-            lambda: client.messages(scoped_key, MODEL, prompt, guardrails=[name], max_tokens=128),
+            lambda: client.messages(scoped_key, MODEL, prompt, guardrails=[name], max_tokens=512),
             _messages_text,
             email=email,
         )
diff --git a/tests/e2e/guardrails/test_streaming_guardrail_e2e.py b/tests/e2e/guardrails/test_streaming_guardrail_e2e.py
index 324dd4107a4..4e4251b9367 100644
--- a/tests/e2e/guardrails/test_streaming_guardrail_e2e.py
+++ b/tests/e2e/guardrails/test_streaming_guardrail_e2e.py
@@ -32,7 +32,7 @@ from lifecycle import ResourceManager
 
 pytestmark = pytest.mark.e2e
 
-MODEL = "gemini-2.5-flash"
+MODEL = "gemini-3.8-flash"
 
 
 class TestBedrockDuringCallStreaming:
diff --git a/tests/e2e/guardrails/test_team_disable_global_guardrail_e2e.py b/tests/e2e/guardrails/test_team_disable_global_guardrail_e2e.py
index 0cc6106abfd..68b1fa99d95 100644
--- a/tests/e2e/guardrails/test_team_disable_global_guardrail_e2e.py
+++ b/tests/e2e/guardrails/test_team_disable_global_guardrail_e2e.py
@@ -20,7 +20,7 @@ from lifecycle import ResourceManager
 
 pytestmark = pytest.mark.e2e
 
-MODEL = "gemini-2.5-flash"
+MODEL = "gemini-3.8-flash"
 
 # A guardrail created via POST /guardrails is registered in-process immediately
 # on the worker that served the create call, but the proxy runs multiple
diff --git a/tests/e2e/guardrails/test_tool_permission_guardrail_e2e.py b/tests/e2e/guardrails/test_tool_permission_guardrail_e2e.py
index d5a6a9d9866..588248c5aba 100644
--- a/tests/e2e/guardrails/test_tool_permission_guardrail_e2e.py
+++ b/tests/e2e/guardrails/test_tool_permission_guardrail_e2e.py
@@ -38,7 +38,7 @@ from models import ChatResponse, ChatTool, ChatToolFunction
 
 pytestmark = pytest.mark.e2e
 
-MODEL = "gemini-2.5-flash"
+MODEL = "gemini-3.8-flash"
 
 #: The one tool the guardrail permits, and one it does not. Both are declared by
 #: the caller in the request body; the guardrail reads them there.
diff --git a/tests/e2e/llm_translation/test_cache_control.py b/tests/e2e/llm_translation/test_cache_control.py
index 7fd51bd6e8c..8a7a315b70b 100644
--- a/tests/e2e/llm_translation/test_cache_control.py
+++ b/tests/e2e/llm_translation/test_cache_control.py
@@ -8,7 +8,7 @@ Each case asserts the feature actually happened, not just a 200. Coverage matrix
   cache-read usage tokens > 0. service_tier is out of scope for Bedrock; AWS
   Bedrock does not expose an OpenAI-style request service tier, so that cell is
   intentionally not covered here.
-- Vertex (gemini-2.5-flash): explicit context caching via ``cache_control``
+- Vertex (gemini-3.8-flash): explicit context caching via ``cache_control``
   with a 5-minute ttl. litellm builds the Vertex cache before the generate
   call, so a never-seen prefix must come back cached on its very first call
   (Gemini's implicit caching cannot hit a cold prefix). That call bills the
@@ -80,7 +80,7 @@ import os
 pytestmark = [pytest.mark.e2e, pytest.mark.provider_live]
 
 BEDROCK_MODEL = "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0"
-VERTEX_MODEL = "vertex_ai/gemini-2.5-flash"
+VERTEX_MODEL = "vertex_ai/gemini-3.8-flash"
 ANTHROPIC_MODEL = "anthropic/claude-haiku-4-5-20251001"
 OPENAI_MODEL = "openai/gpt-5.6"
 VERTEX_CACHE_TTL: Final = "300s"
@@ -380,7 +380,7 @@ class TestCacheControl:
             LiteLLMParamsBody(
                 model=VERTEX_MODEL,
                 vertex_project=os.environ.get("VERTEXAI_PROJECT"),
-                vertex_location="us-central1",
+                vertex_location="global",
                 vertex_credentials=os.environ.get("VERTEXAI_CREDENTIALS"),
             ),
         )
diff --git a/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py b/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
index 1e0a8cbb412..542871a461c 100644
--- a/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
+++ b/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
@@ -45,8 +45,8 @@ from passthrough_client import PassthroughClient
 pytestmark = pytest.mark.e2e
 
 COHERE_BACKEND = "cohere/command-r-08-2024"
-GEMINI_BACKEND = "gemini/gemini-2.5-flash"
-VERTEX_BACKEND: Final = "vertex_ai/gemini-2.5-flash"
+GEMINI_BACKEND = "gemini/gemini-3.8-flash"
+VERTEX_BACKEND: Final = "vertex_ai/gemini-3.8-flash"
 AZURE_OPENAI_BACKEND: Final = "azure/gpt-5.4-nano"
 AZURE_OPENAI_API_VERSION: Final = "v1"
 AZURE_FOUNDRY_BACKEND: Final = "azure_ai/claude-haiku-4-5"
@@ -243,7 +243,7 @@ _PERSON_SCHEMA: dict[str, object] = {
 
 OPENAI_CHAT_MODEL: Final = "gpt-5.5"
 ANTHROPIC_CHAT_MODEL: Final = "claude-haiku-4-5"
-GEMINI_FLASH_MODEL: Final = "gemini-2.5-flash"
+GEMINI_FLASH_MODEL: Final = "gemini-3.8-flash"
 
 CHAT_MODELS: Final = (
     pytest.param(
@@ -449,7 +449,7 @@ class TestVertexChatCompletions:
             LiteLLMParamsBody(
                 model=VERTEX_BACKEND,
                 vertex_project="os.environ/VERTEXAI_PROJECT",
-                vertex_location="us-central1",
+                vertex_location="global",
             ),
         )
         resources.defer(lambda: client.proxy.delete_model(model_id))
@@ -485,7 +485,8 @@ class TestVertexChatCompletions:
                             content=f"Reply with the single word pong. {unique_marker()}",
                         )
                     ],
-                    max_tokens=32,
+                    max_tokens=128,
+                    reasoning_effort="none",
                 ),
             )
         )
@@ -554,7 +555,7 @@ class TestVertexChatCompletions:
 
         response = unwrap(
             client.proxy.chat(
-                key, ChatBody(model=model, messages=_vision_messages(), max_tokens=32, reasoning_effort="none")
+                key, ChatBody(model=model, messages=_vision_messages(), max_tokens=256, reasoning_effort="none")
             )
         )
         _assert_describes_cat(response)
@@ -588,7 +589,8 @@ class TestVertexChatCompletions:
                         content=f"Count from 1 to 5, one number per line. {unique_marker()}",
                     )
                 ],
-                max_tokens=64,
+                max_tokens=128,
+                reasoning_effort="none",
                 stream=True,
             ),
         )
diff --git a/tests/e2e/llm_translation/test_custom_pricing_e2e.py b/tests/e2e/llm_translation/test_custom_pricing_e2e.py
index 6d8cd24648e..5d0cb73a46a 100644
--- a/tests/e2e/llm_translation/test_custom_pricing_e2e.py
+++ b/tests/e2e/llm_translation/test_custom_pricing_e2e.py
@@ -5,7 +5,7 @@ teardown) instead of relying on a statically configured model, so the check is
 self-contained and never inherits pricing another suite or a stale config left on
 the shared proxy. custom-priced-flash sets input/output rates deliberately far
 above the canonical gemini price; the isolation sibling shares the same
-gemini/gemini-2.5-flash backend but sets no override. Three behaviors are checked
+gemini/gemini-3.8-flash backend but sets no override. Three behaviors are checked
 independently:
 
 - billing: a real call's logged cost breakdown charges input and output tokens at
@@ -37,12 +37,12 @@ from models import (
 
 pytestmark = pytest.mark.e2e
 
-BACKEND_MODEL = "gemini/gemini-2.5-flash"
+BACKEND_MODEL = "gemini/gemini-3.8-flash"
 GEMINI_API_KEY = "os.environ/GEMINI_API_KEY"
-# Deliberately ~100x above canonical gemini-2.5-flash (input 3e-7 / output 2.5e-6)
+# Deliberately ~100x above canonical gemini-3.8-flash (input 7.5e-7 / output 3.75e-6)
 # so an override that is ignored or under-applied bills at the base rate and fails.
-CUSTOM_INPUT_RATE = 5e-05
-CUSTOM_OUTPUT_RATE = 1e-04
+CUSTOM_INPUT_RATE = 7.5e-05
+CUSTOM_OUTPUT_RATE = 3.75e-04
 
 
 class _CostBreakdown(BaseModel):
@@ -78,7 +78,7 @@ def _provision(
     input_cost_per_token: float | None,
     output_cost_per_token: float | None,
 ) -> str:
-    """Register a fresh gemini/gemini-2.5-flash deployment (deleted on teardown) and
+    """Register a fresh gemini/gemini-3.8-flash deployment (deleted on teardown) and
     return its model name. With the cost fields set the deployment carries a custom
     pricing override; with them None it is a plain sibling on the same backend. The
     marker keeps the name unique so concurrent runs on the shared proxy never
@@ -251,7 +251,7 @@ class TestCustomPricing:
         assert sibling_entry is not None, f"{sibling} absent from /model/info"
 
         # custom-priced-flash overrides pricing; the sibling shares the same
-        # gemini/gemini-2.5-flash backend but sets no override, so it must keep its
+        # gemini/gemini-3.8-flash backend but sets no override, so it must keep its
         # own price. Equal rates mean the override leaked into the shared cost map.
         assert (
             sibling_entry.model_info.input_cost_per_token
diff --git a/tests/e2e/llm_translation/test_google_native_e2e.py b/tests/e2e/llm_translation/test_google_native_e2e.py
index b18683bd4ad..65a341dbc2a 100644
--- a/tests/e2e/llm_translation/test_google_native_e2e.py
+++ b/tests/e2e/llm_translation/test_google_native_e2e.py
@@ -21,7 +21,7 @@ from pydantic import BaseModel
 
 pytestmark = pytest.mark.e2e
 
-UPSTREAM_MODEL = "gemini/gemini-2.5-flash"
+UPSTREAM_MODEL = "gemini/gemini-3.8-flash"
 
 
 class _GenerateContentPart(BaseModel):
diff --git a/tests/e2e/llm_translation/test_passthrough_e2e.py b/tests/e2e/llm_translation/test_passthrough_e2e.py
index 52d1c10d829..3eaeef6bddd 100644
--- a/tests/e2e/llm_translation/test_passthrough_e2e.py
+++ b/tests/e2e/llm_translation/test_passthrough_e2e.py
@@ -5,7 +5,7 @@ and verifies the proxy still logged a costed SpendLogs row
 (call_type="pass_through_endpoint"), correlated by the id the caller was served:
 the x-litellm-call-id header on gemini, the `msg_...` message id on anthropic.
 
-Covered: gemini ("gemini-2.5-flash") + anthropic ("claude-haiku-4-5"), streaming +
+Covered: gemini ("gemini-3.8-flash") + anthropic ("claude-haiku-4-5"), streaming +
 non-streaming, plus native tool calls. See LLM_TRANSLATION_COVERAGE_MATRIX.md.
 
 A passthrough call returning non-2xx fails hard (never a skip); once it returns
@@ -32,7 +32,7 @@ from passthrough_client import (
     completed_responses_object,
 )
 
-GEMINI_MODEL: Final = "gemini-2.5-flash"
+GEMINI_MODEL: Final = "gemini-3.8-flash"
 ANTHROPIC_PASSTHROUGH_MODEL: Final = "claude-haiku-4-5"
 EMBEDDING_MODEL = "text-embedding-3-small"
 REALTIME_MODEL = "gpt-realtime-2"
@@ -291,7 +291,7 @@ class TestPassthroughModelAllowlist:
 
         result = client.anthropic_message(key, ANTHROPIC_PASSTHROUGH_MODEL, f"say hi {unique_marker()}")
         assert result.status_code == 403, (
-            "a key restricted to gemini-2.5-flash must be denied a claude passthrough call, "
+            "a key restricted to gemini-3.8-flash must be denied a claude passthrough call, "
             f"got {result.status_code}: {result.body[:300]}"
         )
 
diff --git a/tests/e2e/llm_translation/test_responses_e2e.py b/tests/e2e/llm_translation/test_responses_e2e.py
index a0f185c4077..c6cdf8e50de 100644
--- a/tests/e2e/llm_translation/test_responses_e2e.py
+++ b/tests/e2e/llm_translation/test_responses_e2e.py
@@ -56,8 +56,8 @@ OPENAI_MINI_BACKEND: Final = "openai/gpt-4o-mini"
 OPENAI_VISION_BACKEND: Final = "openai/gpt-4o"
 ANTHROPIC_BACKEND: Final = "anthropic/claude-haiku-4-5"
 BEDROCK_CONVERSE_BACKEND: Final = "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0"
-VERTEX_BACKEND: Final = "vertex_ai/gemini-2.5-flash"
-GEMINI_BACKEND: Final = "gemini/gemini-2.5-flash"
+VERTEX_BACKEND: Final = "vertex_ai/gemini-3.8-flash"
+GEMINI_BACKEND: Final = "gemini/gemini-3.8-flash"
 OPENAI_RESPONSES_BACKEND: Final = "openai/gpt-5.5"
 INSTRUCTIONS = "You are a helpful assistant"
 CAT_IMAGE_URL = "https://upload.wikimedia.org/wikipedia/commons/3/3a/Cat03.jpg"
@@ -128,7 +128,7 @@ def _vertex_params() -> LiteLLMParamsBody:
     return LiteLLMParamsBody(
         model=VERTEX_BACKEND,
         vertex_project="os.environ/VERTEXAI_PROJECT",
-        vertex_location="us-central1",
+        vertex_location="global",
     )
 
 
diff --git a/tests/e2e/llm_translation/test_token_counter_gemini_contents_e2e.py b/tests/e2e/llm_translation/test_token_counter_gemini_contents_e2e.py
index d50a7bae9d1..737c8bf2602 100644
--- a/tests/e2e/llm_translation/test_token_counter_gemini_contents_e2e.py
+++ b/tests/e2e/llm_translation/test_token_counter_gemini_contents_e2e.py
@@ -20,8 +20,8 @@ from pydantic import BaseModel
 
 pytestmark = pytest.mark.e2e
 
-GEMINI_STUDIO_DEPLOYMENT: Final = "gemini-2.5-flash"
-GEMINI_VERTEX_DEPLOYMENT: Final = "gemini-2.5-flash-vertex"
+GEMINI_STUDIO_DEPLOYMENT: Final = "gemini-3.8-flash"
+GEMINI_VERTEX_DEPLOYMENT: Final = "gemini-3.8-flash-vertex"
 GEMINI_DEPLOYMENTS = (
     pytest.param(
         GEMINI_STUDIO_DEPLOYMENT,
diff --git a/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py b/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py
index 3daf78a31b2..3a70a858e9b 100644
--- a/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py
+++ b/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py
@@ -24,6 +24,7 @@ model, spend > 0), correlated by the x-litellm-call-id header.
 
 import os
 import time
+from typing import Final
 
 import pytest
 from pydantic import BaseModel
@@ -37,10 +38,10 @@ from passthrough_client import PassthroughClient
 
 pytestmark = pytest.mark.e2e
 
-VERTEX_MODEL = "gemini-2.5-flash"
+VERTEX_MODEL = "gemini-3.8-flash"
 # The added deployment's region and the passthrough URL's region are the same constant,
 # so they always agree; the proxy registers passthrough credentials per project+region.
-VERTEX_LOCATION = os.environ.get("VERTEXAI_LOCATION", "us-central1")
+VERTEX_LOCATION: Final = "global"
 
 
 @pytest.fixture(scope="session")
diff --git a/tests/e2e/logging/test_prometheus_cardinality_e2e.py b/tests/e2e/logging/test_prometheus_cardinality_e2e.py
index 597e1e814e3..d03268d9a0d 100644
--- a/tests/e2e/logging/test_prometheus_cardinality_e2e.py
+++ b/tests/e2e/logging/test_prometheus_cardinality_e2e.py
@@ -30,7 +30,7 @@ from logging_client import LoggingClient
 
 pytestmark = pytest.mark.e2e
 
-DRIVER_MODEL = "gemini-2.5-flash"
+DRIVER_MODEL = "gemini-3.8-flash"
 REQUESTS_METRIC = "litellm_requests_metric_total"
 ALIAS_LABEL = "api_key_alias"
 DISTINCT_KEYS = 3
diff --git a/tests/e2e/logging/test_prometheus_queue_time_e2e.py b/tests/e2e/logging/test_prometheus_queue_time_e2e.py
index 9668322b362..f02269d2631 100644
--- a/tests/e2e/logging/test_prometheus_queue_time_e2e.py
+++ b/tests/e2e/logging/test_prometheus_queue_time_e2e.py
@@ -12,7 +12,7 @@ from logging_client import LoggingClient
 
 pytestmark = pytest.mark.e2e
 
-DRIVER_MODEL = "gemini-2.5-flash"
+DRIVER_MODEL = "gemini-3.8-flash"
 QUEUE_TIME_METRIC = "litellm_request_queue_time_seconds"
 ALIAS_LABEL = "api_key_alias"
 
diff --git a/tests/e2e/management/test_key_management_e2e.py b/tests/e2e/management/test_key_management_e2e.py
index 34f912fe604..5ce0d8e49a9 100644
--- a/tests/e2e/management/test_key_management_e2e.py
+++ b/tests/e2e/management/test_key_management_e2e.py
@@ -242,7 +242,7 @@ class TestKeyManagementRoutes:
             client,
             resources,
             KeyGenerateBody(
-                models=["gpt-5.5", "gemini-2.5-flash"],
+                models=["gpt-5.5", "gemini-3.8-flash"],
                 key_alias=alias,
                 tpm_limit=131313,
                 rpm_limit=141414,
@@ -251,8 +251,8 @@ class TestKeyManagementRoutes:
 
         info = client.proxy.key_info(key)
         assert info.key_alias == alias, f"/key/info reports key_alias {info.key_alias!r}, configured {alias!r}"
-        assert info.models == ["gpt-5.5", "gemini-2.5-flash"], (
-            f"/key/info reports models {info.models}, configured ['gpt-5.5', 'gemini-2.5-flash']"
+        assert info.models == ["gpt-5.5", "gemini-3.8-flash"], (
+            f"/key/info reports models {info.models}, configured ['gpt-5.5', 'gemini-3.8-flash']"
         )
         assert info.tpm_limit == 131313, f"/key/info reports tpm_limit {info.tpm_limit}, configured 131313"
         assert info.rpm_limit == 141414, f"/key/info reports rpm_limit {info.rpm_limit}, configured 141414"
@@ -410,7 +410,7 @@ class TestKeyManagementRoutes:
         outcome = client.proxy.transport.send(
             "/key/update",
             headers=client.proxy.transport.bearer(nonadmin),
-            json=KeyUpdateBody(key=target, models=["gemini-2.5-flash"]),
+            json=KeyUpdateBody(key=target, models=["gemini-3.8-flash"]),
         )
         assert outcome.status_code in (401, 403), (
             f"non-admin key POSTing /key/update must be denied 401/403, got {outcome.status_code}: {outcome.body[:300]}"
diff --git a/tests/e2e/management/test_management_e2e.py b/tests/e2e/management/test_management_e2e.py
index c167a1323cb..7f3ceaeca28 100644
--- a/tests/e2e/management/test_management_e2e.py
+++ b/tests/e2e/management/test_management_e2e.py
@@ -48,7 +48,7 @@ from proxy_client import Converged, await_converged
 
 pytestmark = pytest.mark.e2e
 
-GEMINI_MODEL: Final = "gemini-2.5-flash"
+GEMINI_MODEL: Final = "gemini-3.8-flash"
 OPENAI_MODEL: Final = "gpt-5.5"
 REGENERATE_GRACE_PERIOD = "15s"
 REGENERATE_GRACE_SECONDS = 15.0
@@ -155,7 +155,7 @@ class TestKeyRoutes:
         info = client.proxy.key_info(key)
         assert info.key_alias == alias, f"/key/info reports key_alias {info.key_alias!r}, configured {alias!r}"
         assert info.models == [GEMINI_MODEL], (
-            f"/key/info reports models {info.models}, configured ['gemini-2.5-flash']"
+            f"/key/info reports models {info.models}, configured ['gemini-3.8-flash']"
         )
         assert info.tpm_limit == 424242, (
             f"/key/info reports tpm_limit {info.tpm_limit}, configured 424242"
@@ -501,7 +501,7 @@ class TestTeamRoutes:
         info = client.team_info(team_id)
         assert info.team_alias == alias, f"/team/info reports team_alias {info.team_alias!r}, configured {alias!r}"
         assert info.models == [GEMINI_MODEL], (
-            f"/team/info reports models {info.models}, configured ['gemini-2.5-flash']"
+            f"/team/info reports models {info.models}, configured ['gemini-3.8-flash']"
         )
 
         key = _generate_key(client, resources, KeyGenerateBody(team_id=team_id))
@@ -775,7 +775,7 @@ class TestOrganizationRoutes:
             f"/organization/info reports alias {info.organization_alias!r}, configured {alias!r}"
         )
         assert info.models == [GEMINI_MODEL], (
-            f"/organization/info reports models {info.models}, configured ['gemini-2.5-flash']"
+            f"/organization/info reports models {info.models}, configured ['gemini-3.8-flash']"
         )
 
     @pytest.mark.covers("mgmt.organization.update.persists")
diff --git a/tests/e2e/management/test_team_management_e2e.py b/tests/e2e/management/test_team_management_e2e.py
index 905e3cde206..e513ab1e91b 100644
--- a/tests/e2e/management/test_team_management_e2e.py
+++ b/tests/e2e/management/test_team_management_e2e.py
@@ -248,15 +248,15 @@ class TestTeamManagementRoutes:
         self, client: ManagementClient, resources: ResourceManager
     ) -> None:
         alias = f"e2e-team-info-{unique_marker()}"
-        team_id = _create_team(client, resources, alias, ["gemini-2.5-flash"])
+        team_id = _create_team(client, resources, alias, ["gemini-3.8-flash"])
 
         info = _read_team(client, team_id)
         assert info.team_id == team_id, f"/team/info echoed team_id {info.team_id!r}, requested {team_id!r}"
         assert info.team_info.team_alias == alias, (
             f"/team/info reports team_alias {info.team_info.team_alias!r}, configured {alias!r}"
         )
-        assert info.team_info.models == ["gemini-2.5-flash"], (
-            f"/team/info reports models {info.team_info.models}, configured ['gemini-2.5-flash']"
+        assert info.team_info.models == ["gemini-3.8-flash"], (
+            f"/team/info reports models {info.team_info.models}, configured ['gemini-3.8-flash']"
         )
 
     @pytest.mark.covers("mgmt.team.block.persists")
@@ -264,7 +264,7 @@ class TestTeamManagementRoutes:
     def test_block_then_unblock_persists_to_team_info(
         self, client: ManagementClient, resources: ResourceManager
     ) -> None:
-        team_id = _create_team(client, resources, f"e2e-team-block-{unique_marker()}", ["gemini-2.5-flash"])
+        team_id = _create_team(client, resources, f"e2e-team-block-{unique_marker()}", ["gemini-3.8-flash"])
         assert not _read_team(client, team_id).team_info.blocked, "/team/info reports the team blocked before /team/block"
 
         _set_blocked(client, team_id, blocked=True)
@@ -287,7 +287,7 @@ class TestTeamManagementRoutes:
         self, client: ManagementClient, resources: ResourceManager
     ) -> None:
         user_id = _create_user(client, resources, f"e2e-team-mu-{unique_marker()}@example.com")
-        team_id = _create_team(client, resources, f"e2e-team-mu-{unique_marker()}", ["gemini-2.5-flash"])
+        team_id = _create_team(client, resources, f"e2e-team-mu-{unique_marker()}", ["gemini-3.8-flash"])
         client.add_team_member(team_id, user_id)
         assert _member_role(_read_team(client, team_id), user_id) == "user", (
             f"member {user_id} should start as role 'user' after /team/member_add"
@@ -312,7 +312,7 @@ class TestTeamManagementRoutes:
         self, client: ManagementClient, resources: ResourceManager
     ) -> None:
         user_id = _create_user(client, resources, f"e2e-team-md-{unique_marker()}@example.com")
-        team_id = _create_team(client, resources, f"e2e-team-md-{unique_marker()}", ["gemini-2.5-flash"])
+        team_id = _create_team(client, resources, f"e2e-team-md-{unique_marker()}", ["gemini-3.8-flash"])
         client.add_team_member(team_id, user_id)
         assert _member_role(_read_team(client, team_id), user_id) == "user", (
             f"/team/info does not list {user_id} as a member after /team/member_add"
@@ -378,7 +378,7 @@ class TestTeamManagementRoutes:
         user + team, plus a second user id the member could try to add."""
         member_id = _create_user(client, resources, f"e2e-team-fb-{unique_marker()}@example.com")
         other_id = _create_user(client, resources, f"e2e-team-fb-{unique_marker()}@example.com")
-        team_id = _create_team(client, resources, f"e2e-team-fb-{unique_marker()}", ["gemini-2.5-flash"])
+        team_id = _create_team(client, resources, f"e2e-team-fb-{unique_marker()}", ["gemini-3.8-flash"])
         client.add_team_member(team_id, member_id)
         member_key = _generate_key(client, resources, KeyGenerateBody(user_id=member_id, team_id=team_id))
         return member_id, other_id, member_key, team_id
@@ -534,7 +534,7 @@ class TestTeamAdminWithTpmLimitEnabled:
             pytest.param(TeamSettingsChange(rpm_limit=10), id="rpm_limit"),
             pytest.param(TeamSettingsChange(max_budget=0.5), id="max_budget"),
             pytest.param(TeamSettingsChange(team_alias="renamed-by-team-admin"), id="team_alias"),
-            pytest.param(TeamSettingsChange(models=["gemini-2.5-flash"]), id="models"),
+            pytest.param(TeamSettingsChange(models=["gemini-3.8-flash"]), id="models"),
             pytest.param(TeamSettingsChange(budget_duration="1d"), id="budget_duration"),
             pytest.param(TeamSettingsChange(metadata=TeamCustomMetadata(cost_center="team-admin")), id="metadata"),
         ],
diff --git a/tests/e2e/quota_management/budgets/test_model_max_budget_e2e.py b/tests/e2e/quota_management/budgets/test_model_max_budget_e2e.py
index c69b0e232ff..77f779a6efc 100644
--- a/tests/e2e/quota_management/budgets/test_model_max_budget_e2e.py
+++ b/tests/e2e/quota_management/budgets/test_model_max_budget_e2e.py
@@ -20,7 +20,7 @@ from models import ModelBudgetEntry
 pytestmark = pytest.mark.e2e
 
 CAPPED_MODEL = "claude-haiku-4-5"
-FREE_MODEL = "gemini-2.5-flash"
+FREE_MODEL = "gemini-3.8-flash"
 
 
 def _call(client: BudgetClient, key: str, model: str):
diff --git a/tests/e2e/quota_management/spend_tracking/conftest.py b/tests/e2e/quota_management/spend_tracking/conftest.py
index 9c8ffd18144..75ba870137f 100644
--- a/tests/e2e/quota_management/spend_tracking/conftest.py
+++ b/tests/e2e/quota_management/spend_tracking/conftest.py
@@ -32,7 +32,7 @@ def _driver_params(provider_model: str, env_var: str) -> LiteLLMParamsBody:
 
 
 DRIVER_MODELS: tuple[tuple[str, str, str], ...] = (
-    ("gemini-2.5-flash", "gemini/gemini-2.5-flash", "GEMINI_API_KEY"),
+    ("gemini-3.8-flash", "gemini/gemini-3.8-flash", "GEMINI_API_KEY"),
     ("claude-haiku-4-5", "anthropic/claude-haiku-4-5", "ANTHROPIC_API_KEY"),
     ("openai-text-embedding-3-small", "openai/text-embedding-3-small", "OPENAI_API_KEY"),
     ("openai-responses-codex", "openai/gpt-5.3-codex", "OPENAI_API_KEY"),
diff --git a/tests/e2e/quota_management/spend_tracking/test_key_attribution_e2e.py b/tests/e2e/quota_management/spend_tracking/test_key_attribution_e2e.py
index e0c19ea1b6b..1d0878d073a 100644
--- a/tests/e2e/quota_management/spend_tracking/test_key_attribution_e2e.py
+++ b/tests/e2e/quota_management/spend_tracking/test_key_attribution_e2e.py
@@ -55,7 +55,7 @@ from spend_e2e_client import (
 
 pytestmark = pytest.mark.e2e
 
-CHAT_MODEL: Final = "gemini-2.5-flash"
+CHAT_MODEL: Final = "gemini-3.8-flash"
 MESSAGES_MODEL: Final = "claude-haiku-4-5"
 RESPONSES_MODEL: Final = "openai-responses-codex"
 EMBED_MODEL: Final = "openai-text-embedding-3-small"
diff --git a/tests/e2e/quota_management/spend_tracking/test_spend_tracking_e2e.py b/tests/e2e/quota_management/spend_tracking/test_spend_tracking_e2e.py
index c7ef81f826a..d452525ec91 100644
--- a/tests/e2e/quota_management/spend_tracking/test_spend_tracking_e2e.py
+++ b/tests/e2e/quota_management/spend_tracking/test_spend_tracking_e2e.py
@@ -3,7 +3,7 @@
 Run against a proxy started with the gateway config. Coverage rationale:
 SPEND_TRACKING_COVERAGE_MATRIX.md.
 
-Model names are literals from that config: chat tests hit "gemini-2.5-flash",
+Model names are literals from that config: chat tests hit "gemini-3.8-flash",
 embedding tests hit "openai-text-embedding-3-small".
 
 Every test: fresh scoped key (isolation) -> real provider call -> unwrap (hard
@@ -37,7 +37,7 @@ from spend_reconciliation import BACKEND as TRAFFIC_BACKEND
 
 pytestmark = pytest.mark.e2e
 
-GEMINI_MODEL = "gemini-2.5-flash"
+GEMINI_MODEL = "gemini-3.8-flash"
 CLAUDE_MODEL = "claude-haiku-4-5"
 CODEX_MODEL = "openai-responses-codex"
 EMBEDDING_MODEL = "openai-text-embedding-3-small"
@@ -780,7 +780,7 @@ def test_spend_calculate_returns_nonzero_cost(client: SpendClient) -> None:
         GEMINI_MODEL, "estimate the cost of this request"
     )
     assert cost > 0, (
-        "/spend/calculate returned 0 for gemini-2.5-flash; "
+        "/spend/calculate returned 0 for gemini-3.8-flash; "
         "cost map may be missing this model"
     )