fix(ci): use GPT models available via HAI proxy

HAI proxy serves Anthropic on a different path (/anthropic/v1) than
OpenAI (/openai/v1). Until review-bot supports multiple base URLs,
use GPT-5 and GPT-5-mini for both review slots.

.gitea/workflows/ci.yml

@@ -36,7 +36,7 @@ jobs:
           REVIEWER_TOKEN: ${{ secrets.SONNET_REVIEW_TOKEN }}
           LLM_BASE_URL: ${{ secrets.LLM_BASE_URL }}
           LLM_API_KEY: ${{ secrets.LLM_API_KEY }}
-          LLM_MODEL: "anthropic--claude-4.6-sonnet"
+          LLM_MODEL: "gpt-5"
           CONVENTIONS_FILE: "CONVENTIONS.md"
         run: ./review-bot
       - name: Run GPT Review
@@ -47,6 +47,6 @@ jobs:
           REVIEWER_TOKEN: ${{ secrets.GPT_REVIEW_TOKEN }}
           LLM_BASE_URL: ${{ secrets.LLM_BASE_URL }}
           LLM_API_KEY: ${{ secrets.LLM_API_KEY }}
-          LLM_MODEL: "sap-ai-opus-latest-openai/gpt-5"
+          LLM_MODEL: "gpt-5-mini"
           CONVENTIONS_FILE: "CONVENTIONS.md"
         run: ./review-bot