You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
First: awesome work on the SDK — we’re using Strands in production-style agent workflows and want to standardize on Gemini 3 for some workloads.
What I’m trying to do
Use the Gemini provider with:
model_id="gemini-3-pro-preview"
Tool calling enabled (regular Strands tools / function calling)
Current behavior
After the model performs a tool call, the next request fails with:
google.genai.errors.ClientError: 400 Bad Request
status: INVALID_ARGUMENT
This appears tied to Gemini 3’s requirement to pass back the “thought signature” (sometimes described as a mandatory signature) exactly as received when continuing the conversation after a tool call.
Tool calling should work on Gemini 3 models the same way it works on Gemini 2.x / 2.5.x models (or at least fail with a clearer, actionable error + workaround).
Questions (roadmap + implementation)
Is Gemini 3 officially supported / planned?
If yes: what’s the target version/milestone?
Will Strands handle “thought signature” propagation automatically for Gemini 3 tool-calling flows?
e.g., persist and return the signature fields from model responses in the next request payloads.
If there’s a recommended workaround today (flags, provider settings, disabling a feature, etc.), could you share a minimal example?
Would you accept a PR if someone implements the signature propagation?
If yes: any pointers to the best place in the provider code where this should live?
Minimal repro (pseudo)
fromstrandsimportAgentfromstrands.models.geminiimportGeminiModelmodel=GeminiModel(model_id="gemini-3-pro-preview")
agent=Agent(model=model, tools=[...])
agent("Do something that triggers a tool call")
# Next continuation after tool result → INVALID_ARGUMENTHappytoprovide:
afullrunnablereprorepodebuglogs/requesttraces (withsecretsremoved)
Thanksinadvance — Gemini3supportwouldbeahugeunlockforteamsstandardizingonGoogle’slatestmodels.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
-
Hi Strands team 👋
First: awesome work on the SDK — we’re using Strands in production-style agent workflows and want to standardize on Gemini 3 for some workloads.
What I’m trying to do
Use the Gemini provider with:
model_id="gemini-3-pro-preview"Current behavior
After the model performs a tool call, the next request fails with:
google.genai.errors.ClientError: 400 Bad Requeststatus: INVALID_ARGUMENTThis appears tied to Gemini 3’s requirement to pass back the “thought signature” (sometimes described as a mandatory signature) exactly as received when continuing the conversation after a tool call.
Related bug report:
Expected behavior
Tool calling should work on Gemini 3 models the same way it works on Gemini 2.x / 2.5.x models (or at least fail with a clearer, actionable error + workaround).
Questions (roadmap + implementation)
Is Gemini 3 officially supported / planned?
Will Strands handle “thought signature” propagation automatically for Gemini 3 tool-calling flows?
If there’s a recommended workaround today (flags, provider settings, disabling a feature, etc.), could you share a minimal example?
Would you accept a PR if someone implements the signature propagation?
Minimal repro (pseudo)
Beta Was this translation helpful? Give feedback.
All reactions