diff --git a/litellm/google_genai/adapters/transformation.py b/litellm/google_genai/adapters/transformation.py
index e1085dd7857..0363a6d3d1c 100644
--- a/litellm/google_genai/adapters/transformation.py
+++ b/litellm/google_genai/adapters/transformation.py
@@ -628,10 +628,14 @@ class GoogleGenAIAdapter:
     ) -> list[AllMessageValues]:
         """Transform Google GenAI contents to OpenAI messages format"""
         system_parts: Final = system_instruction.get("parts", ()) if system_instruction else ()
+        system_texts: Final = tuple(part["text"] for part in system_parts if "text" in part)
+        system_content: Final = (
+            system_texts[0]
+            if len(system_texts) == 1
+            else [ChatCompletionTextObject(type="text", text=text) for text in system_texts]
+        )
         system_messages: Final = (
-            (ChatCompletionSystemMessage(role="system", content=system_parts[0]["text"]),)
-            if system_parts and "text" in system_parts[0]
-            else ()
+            (ChatCompletionSystemMessage(role="system", content=system_content),) if system_texts else ()
         )
         return [*system_messages, *chain.from_iterable(_genai_content_messages(content) for content in contents)]
 
diff --git a/litellm/proxy/spend_tracking/budget_reservation.py b/litellm/proxy/spend_tracking/budget_reservation.py
index cfc865a84d1..5cada93bd33 100644
--- a/litellm/proxy/spend_tracking/budget_reservation.py
+++ b/litellm/proxy/spend_tracking/budget_reservation.py
@@ -16,6 +16,7 @@ from pydantic import ConfigDict, TypeAdapter
 import litellm
 from litellm._internal_context import with_service_target
 from litellm._logging import verbose_proxy_logger
+from litellm.litellm_core_utils.core_helpers import as_str_mapping
 from litellm.litellm_core_utils.duration_parser import duration_in_seconds
 from litellm.litellm_core_utils.llm_cost_calc.tiered_pricing import select_tier_for_input, tier_rate
 from litellm.proxy._types import (
@@ -1652,24 +1653,27 @@ def _estimate_output_tokens(
 _OUTPUT_TOKEN_FIELDS: Final = ("max_completion_tokens", "max_tokens", "max_output_tokens")
 
 
+def _nested_field(request_body: Mapping[str, object], config_field: str, field: str) -> object:
+    config: Final = as_str_mapping(request_body.get(config_field))
+    return config.get(field) if config is not None else None
+
+
 def _requested_output_tokens(request_body: Mapping[str, object]) -> int | None:
-    inference_config: Final = request_body.get("inferenceConfig")
-    generation_config: Final = request_body.get("generationConfig")
     candidates: Final = (
         *(request_body.get(field) for field in _OUTPUT_TOKEN_FIELDS),
-        inference_config.get("maxTokens") if isinstance(inference_config, Mapping) else None,
-        generation_config.get("maxOutputTokens") if isinstance(generation_config, Mapping) else None,
+        _nested_field(request_body, "inferenceConfig", "maxTokens"),
+        _nested_field(request_body, "generationConfig", "maxOutputTokens"),
     )
     return next((tokens for tokens in map(_to_int, candidates) if tokens is not None), None)
 
 
-def _get_output_multiplier(request_body: dict) -> int:
-    output_multiplier = 1
-    for key in ("n", "best_of"):
-        value = _to_int(request_body.get(key))
-        if value is not None:
-            output_multiplier = max(output_multiplier, value)
-    return output_multiplier
+def _get_output_multiplier(request_body: Mapping[str, object]) -> int:
+    candidates: Final = (
+        request_body.get("n"),
+        request_body.get("best_of"),
+        _nested_field(request_body, "generationConfig", "candidateCount"),
+    )
+    return max((1, *(count for count in map(_to_int, candidates) if count is not None)))
 
 
 def _is_input_only_route(route: str) -> bool:
diff --git a/tests/unit/google_genai/test_google_genai_adapter.py b/tests/unit/google_genai/test_google_genai_adapter.py
index 0292e58c029..7d1e8e2caa4 100644
--- a/tests/unit/google_genai/test_google_genai_adapter.py
+++ b/tests/unit/google_genai/test_google_genai_adapter.py
@@ -182,6 +182,27 @@ def test_native_function_call_preserves_arguments_and_matching_response(args: ob
     ]
 
 
+@pytest.mark.parametrize(
+    ("system_texts", "expected_content"),
+    [
+        (("Answer tersely",), "Answer tersely"),
+        (
+            ("Answer tersely", "Use British spelling"),
+            [{"type": "text", "text": "Answer tersely"}, {"type": "text", "text": "Use British spelling"}],
+        ),
+    ],
+)
+def test_every_system_instruction_text_part_reaches_the_system_message(
+    system_texts: tuple[str, ...], expected_content: str | list[dict[str, str]]
+) -> None:
+    request: Final = GoogleGenAIAdapter().translate_generate_content_to_completion(
+        model="gemini/gemini-3.8-flash",
+        contents=[{"role": "user", "parts": [{"text": "hi"}]}],
+        systemInstruction={"parts": [{"text": text} for text in system_texts]},
+    )
+    assert request["messages"][0] == {"role": "system", "content": expected_content}
+
+
 def test_adapter_import():
     """Test that the adapter can be imported successfully"""
     from litellm.google_genai.adapters.handler import GenerateContentToCompletionHandler
diff --git a/tests/unit/proxy/spend_tracking/test_budget_reservation.py b/tests/unit/proxy/spend_tracking/test_budget_reservation.py
index f1ae7bcb5cd..78431304198 100644
--- a/tests/unit/proxy/spend_tracking/test_budget_reservation.py
+++ b/tests/unit/proxy/spend_tracking/test_budget_reservation.py
@@ -177,7 +177,7 @@ def test_bedrock_converse_body_reserves_the_prompt_not_the_context_window():
 
 
 GEMINI_PROMPT: Final = "Say hello in five words"
-GEMINI_SYSTEM: Final = "Answer tersely"
+GEMINI_SYSTEM_PARTS: Final = ("Answer tersely", "Use British spelling")
 
 
 @pytest.mark.parametrize(
@@ -190,14 +190,22 @@ GEMINI_SYSTEM: Final = "Answer tersely"
 )
 @pytest.mark.parametrize(
     ("native_output_config", "chat_output_limit"),
-    (({"generationConfig": {"maxOutputTokens": 5}}, {"max_tokens": 5}), ({}, {})),
+    (
+        ({"generationConfig": {"maxOutputTokens": 5}}, {"max_tokens": 5}),
+        ({"generationConfig": {"maxOutputTokens": 5, "candidateCount": 3}}, {"max_tokens": 5, "n": 3}),
+        ({}, {}),
+    ),
 )
+@pytest.mark.parametrize("system_texts", (GEMINI_SYSTEM_PARTS[:1], GEMINI_SYSTEM_PARTS))
 def test_gemini_generate_content_body_reserves_the_prompt_not_the_context_window(
-    route: str, native_output_config: dict[str, object], chat_output_limit: dict[str, object]
+    route: str,
+    native_output_config: dict[str, object],
+    chat_output_limit: dict[str, object],
+    system_texts: tuple[str, ...],
 ) -> None:
     native_cost: Final = estimate_request_max_cost(
         request_body={
-            "systemInstruction": {"parts": [{"text": GEMINI_SYSTEM}]},
+            "systemInstruction": {"parts": [{"text": text} for text in system_texts]},
             "contents": [{"role": "user", "parts": [{"text": GEMINI_PROMPT}]}],
             **native_output_config,
         },
@@ -209,7 +217,7 @@ def test_gemini_generate_content_body_reserves_the_prompt_not_the_context_window
         request_body={
             "model": "gemini-3.8-flash",
             "messages": [
-                {"role": "system", "content": GEMINI_SYSTEM},
+                {"role": "system", "content": [{"type": "text", "text": text} for text in system_texts]},
                 {"role": "user", "content": GEMINI_PROMPT},
             ],
             **chat_output_limit,