Disable Gemini thinking budget in extraction backend

gemini-2.5-flash thinks by default and spent ~3.8k of the 4k output budget on reasoning, hitting MAX_TOKENS with a truncated JSON body -> 0 claims parsed. Set thinking_budget=0 so the full budget goes to the answer (mirrors the local path's enable_thinking=False). On the validation chunk this went from 0 -> 11 claims.
2026-06-15 22:28:12 -05:00
parent 5deffddb17
commit e8d50efdf4
1 changed files with 4 additions and 0 deletions
@@ -50,6 +50,10 @@ class GeminiBackend:
                temperature=0,
                max_output_tokens=max_tokens,
                response_mime_type="application/json",
+                # Gemini 2.5 thinks by default and spends the output budget on reasoning tokens —
+                # it hit MAX_TOKENS with ~3.8k thoughts and a truncated JSON body (0 claims parsed).
+                # Extraction is deterministic, no-CoT (mirrors the local path's enable_thinking=False).
+                thinking_config=types.ThinkingConfig(thinking_budget=0),
            ),
        )
        return resp.text or "{}"