Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading to OpenClaw 2026.6.5, a configured default model of google/gemini-3.1-pro-preview was retained in config but was no longer recognized by openclaw infer model inspect or openclaw infer model run, causing Gemini-backed agent responses to fail until the model configuration was manually repaired.

Steps to reproduce

1. Upgrade to OpenClaw 2026.6.5 with a configuration containing google/gemini-3.1-pro-preview as the default model.
2. Run openclaw infer model auth status and observe google/gemini-3.1-pro-preview as the resolved default model.
3. Run openclaw infer model inspect --model google/gemini-3.1-pro-preview.
4. Observe Model not found.
5. Run openclaw infer model run --model google/gemini-3.1-pro-preview --prompt "Reply with OK only.".
6. Observe Unknown model.
7. Run openclaw infer model list | grep -i '"provider":"google"'.
8. Observe that only google/gemini-2.5-pro is present in the native Google model catalog.

Expected behavior

A configured default model should either remain runnable after upgrade or be automatically migrated, removed, or flagged with a clear validation error if the model is no longer supported by the runtime catalog.

Actual behavior

google/gemini-3.1-pro-preview remained configured as the default and resolved default model, but openclaw infer model inspect --model google/gemini-3.1-pro-preview returned Model not found and openclaw infer model run --model google/gemini-3.1-pro-preview --prompt "Reply with OK only." returned Unknown model; openclaw infer model list showed only google/gemini-2.5-pro in the native Google model catalog.

OpenClaw version

2026.6.5

Operating system

Linux Mint 22.3 Zena, based on Ubuntu 24.04 LTS

Install method

npm global

Model

google/gemini-3.1-pro-preview

Provider / routing chain

google

Additional provider/model setup details

Configured default model:

google/gemini-3.1-pro-preview

Observed after upgrade:

• openclaw infer model inspect --model google/gemini-3.1-pro-preview returned Model not found.
• openclaw infer model run --model google/gemini-3.1-pro-preview --prompt "Reply with OK only." returned Unknown model.
• openclaw infer model list | grep -i '"provider":"google"' showed only:

{"id":"gemini-2.5-pro","name":"Gemini 2.5 Pro","provider":"google"}

The system was restored by changing the default model to google/gemini-2.5-pro.

Logs, screenshots, and evidence

### Auth status


{
  "defaultModel": "google/gemini-3.1-pro-preview",
  "resolvedDefault": "google/gemini-3.1-pro-preview"
}


### Model inspection


$ openclaw infer model inspect --model google/gemini-3.1-pro-preview

Model not found


### Model execution


$ openclaw infer model run --model google/gemini-3.1-pro-preview --prompt "Reply with OK only."

Unknown model


### Native Google model catalog


$ openclaw infer model list | grep -i '"provider":"google"\|"gemini-3"\|"gemini-2.5"'

{"id":"gemini-2.5-pro","name":"Gemini 2.5 Pro","provider":"google","contextWindow":200000,"reasoning":false,"input":["text"]}

Impact and severity

Affected: Users with configurations that still reference google/gemini-3.1-pro-preview after upgrading to OpenClaw 2026.6.5.

Severity: High. The configured default model cannot be inspected or executed, which can prevent agent responses until the model configuration is manually corrected.

Frequency: Observed consistently in this environment.

Consequence: Gemini-backed agents may stop responding because the configured default model is not runnable by the inference runtime.

Additional information

The configuration contained google/gemini-3.1-pro-preview as both the configured default model and the resolved default model.

Observed runtime behavior:

• openclaw infer model inspect --model google/gemini-3.1-pro-preview returned Model not found.
• openclaw infer model run --model google/gemini-3.1-pro-preview --prompt "Reply with OK only." returned Unknown model.
• openclaw infer model list showed only google/gemini-2.5-pro in the native Google provider catalog.

Workaround: change the default model to google/gemini-2.5-pro.