Skip to content

fix: filter Gemini thinking parts from user-facing message chain #7196

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Soulter merged 1 commit into from
Apr 3, 2026
Merged

fix: filter Gemini thinking parts from user-facing message chain #7196

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 9 additions & 10 deletions astrbot/core/provider/sources/gemini_source.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,15 +241,10 @@ async def _prepare_query_config(
thinking_config = types.ThinkingConfig(
thinking_budget=thinking_budget,
)
elif model_name in [
"gemini-3-pro",
"gemini-3-pro-preview",
"gemini-3-flash",
"gemini-3-flash-preview",
"gemini-3-flash-lite",
"gemini-3-flash-lite-preview",
]:
# The thinkingLevel parameter, recommended for Gemini 3 models and onwards
elif any(model_name.startswith(p) for p in ("gemini-3-", "gemini-3.")):
# The thinkingLevel parameter, recommended for Gemini 3 models and onwards.
# Use prefix match so new variants (3.1, 3-flash-lite-preview, etc.) are
# covered without needing to keep an exhaustive list up to date.
# Gemini 2.5 series models don't support thinkingLevel; use thinkingBudget instead.
thinking_level = self.provider_config.get("gm_thinking_config", {}).get(
"level", "HIGH"
Expand Down Expand Up @@ -517,7 +512,11 @@ def _process_content_parts(
):
chain.append(Comp.Plain("这是图片"))
for part in result_parts:
if part.text:
# Skip thinking parts — their text is already captured via
# _extract_reasoning_content above. Including them here would
# leak the model's internal reasoning into the user-facing message,
# which also causes duplicate/triple replies on some platforms.
if part.text and not part.thought:
chain.append(Comp.Plain(part.text))

if (
Expand Down
Loading