diff --git a/apps/api/src/core/config.py b/apps/api/src/core/config.py
index a81e8a2b..26b97a56 100644
--- a/apps/api/src/core/config.py
+++ b/apps/api/src/core/config.py
@@ -370,9 +370,10 @@ class Settings(BaseSettings):
             )
         return v
 
-    # 2026-04-25 Claude Engineer-C (P1.1): Ollama 健康檢測推理測試模型
+    # 2026-05-05 Codex: health inference must stay on alert-fast model; qwen2.5
+    # keeps reloading a 7B model on CPU-only GCP and slows incident fallback.
     OLLAMA_HEALTH_CHECK_MODEL: str = Field(
-        default="qwen2.5:7b-instruct",
+        default="gemma3:4b",
         description="OllamaHealthMonitor 推理測試使用模型（P1.1）",
     )
     # 2026-04-12 ogt: 心跳必須確認載入的 Ollama 模型清單
diff --git a/docs/LOGBOOK.md b/docs/LOGBOOK.md
index 45aa2727..3a738b70 100644
--- a/docs/LOGBOOK.md
+++ b/docs/LOGBOOK.md
@@ -3201,6 +3201,7 @@ bash scripts/ops/ollama-topology-check.sh
 - `interactive` / `healthcheck` / `alert_fast` 保持 GCP-A 優先
 - `code_review` / `rag` / `embedding` / `deep_rca` / `image_analysis` / `hermes` 改為 111 優先
 - 111 不可用時才回 GCP-B，避免 GCP-A/B 在告警 canary 期間被 7B/14B/32B 模型污染
+- `OLLAMA_HEALTH_CHECK_MODEL` 改為 `gemma3:4b`，避免 health probe 自己把 `qwen2.5:7b-instruct` 載入 GCP-A
 
 驗證：
 
diff --git a/k8s/awoooi-prod/06-deployment-api.yaml b/k8s/awoooi-prod/06-deployment-api.yaml
index 41799f72..e5aa26e7 100644
--- a/k8s/awoooi-prod/06-deployment-api.yaml
+++ b/k8s/awoooi-prod/06-deployment-api.yaml
@@ -79,6 +79,8 @@ spec:
               value: "true"  # 告警診斷強制先走 GCP-A → GCP-B → 111
             - name: ALERT_OLLAMA_MODEL
               value: "gemma3:4b"  # 2026-05-05 Codex: qwen3:14b 告警 JSON prompt 會拖到 504
+            - name: OLLAMA_HEALTH_CHECK_MODEL
+              value: "gemma3:4b"  # 2026-05-05 Codex: 避免 health probe 載入 qwen2.5 7B 污染 GCP alert lane
             - name: OPENCLAW_DEFAULT_MODEL
               value: "qwen2.5:7b-instruct"
             - name: OPENCLAW_TIMEOUT