Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
When a non-default model is selected via /model for a session override, the session-selected model still appears in the Fallbacks list shown by /status. This means if the current session model fails, the first fallback candidate could be the same already-selected model, creating an ineffective fallback chain.
Steps to reproduce
Confirm current model is config default (google/gemini-3-flash-preview)
Use /model to select any non-default model (e.g. google/gemini-3.1-flash-lite)
Run /status
Observe Fallbacks list still includes the session-selected model
Expected behavior
When a session model override is active, the Fallbacks list should exclude the currently selected session model to avoid self-referential fallback routing.
Actual behavior
Session-selected model appears as an entry in the Fallbacks list:
Session selected: google/gemini-3.1-flash-lite
Fallbacks: google/gemini-3.1-flash-lite, google/gemini-2.5-flash, google/gemini-3.1-pro-preview
This affects all non-default model selections. Confirmed with:
google/gemini-3.1-flash-lite → appears in Fallbacks
google/gemini-2.5-flash → appears in Fallbacks
Selecting the config default (google/gemini-3-flash-preview) works correctly — no session override warning and Fallbacks list is normal.
OpenClaw version
2026.5.26 (10ad3aa)
Operating system
Rocky Linux, Node 24.14.1
Install method
npm global
Model
google/gemini-3-flash-preview (config default)
Provider / routing chain
Telegram → OpenClaw Gateway → Google Generative AI
Additional provider/model setup details
"model": {
"primary": "google/gemini-3-flash-preview",
"fallbacks": [
"google/gemini-3.1-flash-lite",
"google/gemini-2.5-flash",
"google/gemini-3.1-pro-preview"
]
}
Logs, screenshots, and evidence
/status command output shows the session-selected model appearing in the Fallbacks list. Screenshots attached in follow-up comment.
Note: Gateway log (/tmp/openclaw/openclaw-2026-05-29.log) contains no entries for /model slash command execution — the command appears to be handled entirely within the gateway without generating log output, making log-based diagnosis impossible.
Impact and severity
Severity: High — When a session model override is active and that model fails, the fallback chain will attempt the same failed model again as the first fallback candidate before trying genuinely different models. This wastes API calls and delays recovery, potentially causing cascading failures across the entire fallback chain.
Frequency: Reproducible every time a non-default model is selected via /model. Affects all Google provider models confirmed with google/gemini-3.1-flash-lite and google/gemini-2.5-flash.
Affected scope: All users using /model session override with Google provider. Any non-default model selection triggers this behavior.
Additional information
Last known good version: NOT_ENOUGH_INFO — behavior may have existed before 2026.5.26 but was noticed after upgrading. Selecting config default model correctly shows no session override and normal fallback behavior.
Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
When a non-default model is selected via /model for a session override, the session-selected model still appears in the Fallbacks list shown by /status. This means if the current session model fails, the first fallback candidate could be the same already-selected model, creating an ineffective fallback chain.
Steps to reproduce
Confirm current model is config default (google/gemini-3-flash-preview)
Use /model to select any non-default model (e.g. google/gemini-3.1-flash-lite)
Run /status
Observe Fallbacks list still includes the session-selected model
Expected behavior
When a session model override is active, the Fallbacks list should exclude the currently selected session model to avoid self-referential fallback routing.
Actual behavior
Session-selected model appears as an entry in the Fallbacks list:
Session selected: google/gemini-3.1-flash-lite
Fallbacks: google/gemini-3.1-flash-lite, google/gemini-2.5-flash, google/gemini-3.1-pro-preview
This affects all non-default model selections. Confirmed with:
google/gemini-3.1-flash-lite → appears in Fallbacks
google/gemini-2.5-flash → appears in Fallbacks
Selecting the config default (google/gemini-3-flash-preview) works correctly — no session override warning and Fallbacks list is normal.
OpenClaw version
2026.5.26 (10ad3aa)
Operating system
Rocky Linux, Node 24.14.1
Install method
npm global
Model
google/gemini-3-flash-preview (config default)
Provider / routing chain
Telegram → OpenClaw Gateway → Google Generative AI
Additional provider/model setup details
"model": {
"primary": "google/gemini-3-flash-preview",
"fallbacks": [
"google/gemini-3.1-flash-lite",
"google/gemini-2.5-flash",
"google/gemini-3.1-pro-preview"
]
}
Logs, screenshots, and evidence
Impact and severity
Severity: High — When a session model override is active and that model fails, the fallback chain will attempt the same failed model again as the first fallback candidate before trying genuinely different models. This wastes API calls and delays recovery, potentially causing cascading failures across the entire fallback chain.
Frequency: Reproducible every time a non-default model is selected via /model. Affects all Google provider models confirmed with google/gemini-3.1-flash-lite and google/gemini-2.5-flash.
Affected scope: All users using /model session override with Google provider. Any non-default model selection triggers this behavior.
Additional information
Last known good version: NOT_ENOUGH_INFO — behavior may have existed before 2026.5.26 but was noticed after upgrading. Selecting config default model correctly shows no session override and normal fallback behavior.