Repository navigation
Gemini 3 is not supported via LiteLLM #2109
Description
Please read this first
- Have you read the custom model provider docs, including the 'Common issues' section? Model provider docs
- Have you searched for related issues? Others may have faced similar issues.
Describe the question
Can not use Gemini 3 model (gemini/gemini-3-pro-preview) via LiteLLM when there is a tool call, with or without model_settings=ModelSettings(reasoning=Reasoning(effort="low")).
Debug information
- Agents SDK version: v0.6.1
- Python version: 3.13
Repro steps
from agents import Agent, ModelSettings, Runner, function_tool
from agents.extensions.models.litellm_model import LitellmModel
from openai.types.shared import Reasoning
@function_tool
def echo_tool(input: str) -> str:
return input
model = LitellmModel(
model="gemini/gemini-3-pro-preview",
api_key="[API_KEY]",
)
agent = Agent(
name="Assistant",
instructions="You are a helpful assistant",
model=model,
tools=[echo_tool],
# model_settings=ModelSettings(reasoning=Reasoning(effort="low")),
)
result = Runner.run_sync(
agent, "Can you use the echo_tool to echo the message 'Hello, world!'?"
)
print(result.final_output)It prints
Function call is missing a thought_signature in functionCall parts. This is required for tools to work correctly, and missing thought_signature may lead to degraded model performance. Additional data, function call `default_api:echo_tool` , position 2. please refer to https://ai.google.dev/gemini-api/docs/thought-signatures for more details.
Activity
ihower commentedon Nov 20, 2025
This will need a proper LiteLLM release that supports Gemini 3 Pro first.
LiteLLM has an early implementation here: BerriAI/litellm#16719 but there is no PyPI release for it yet.
Once LiteLLM ships a release with this support, I can provide a hotfix PR as soon as possible.
kkwhale7 commentedon Nov 24, 2025
BerriAI/litellm#16780
I noticed that LiteLLM already appears to support Gemini 3
lawalkeyd commentedon Nov 24, 2025
Released in 1.80.5
ihower commentedon Nov 24, 2025
@kkwhale7 @lawalkeyd @Nifury Quick update on the progress: I’ve finished the first implementation on my fork branch ihower#2, and it works on my side.
If you want to try it early, you can install it with:
And please make sure your LiteLLM version is upgraded to 1.80.5.
I’m now preparing the formal pull request and expect to submit it later this week.
ralliesai commentedon Dec 4, 2025
@ihower thanks so much for working on this. Any updates here?
ZILECAO commentedon Jan 7, 2026
goat @ihower , thank you
seratch commentedon Jan 7, 2026
resolved by #2158