Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

+1 — hit this exact bug in production today running LangExtract 1.5.0 with gemini-2.5-flash.

Symptom: Some lx.extract() calls returned 0 extractions with Skipping chunk: parse error: Failed to parse JSON content: Unterminated string starting at: line 1 column 698 (char 697). HTTP 200 from Gemini, but the JSON response body was truncated mid-string. Same prompt sometimes succeeded, sometimes truncated — i.e. thinking budget was the variable.

Confirmed: passing thinking_config via language_model_params was silently dropped by _API_CONFIG_KEYS filtering (line 287 + line 405), exactly as described in #319.

Workaround we shipped: import-time monkeypatch that adds 'thinking_config' to _API_CONFIG_KEYS, plus language_model_params={'max_output_tokens': 16384, 'thinking_config': ThinkingConfig(thinking_budget=0)}. The patch itself works (we can see thinking_config forwarded), but in our specific workload some complex prompts still hit truncation even with thinking_budget=0 — so we ultimately switched our LangExtract path to gemini-2.5-flash-lite (thinking off by default) while keeping gemini-2.5-flash for direct SDK calls.

The single-line allowlist addition in this PR is the right fix and would have saved us a day of debugging. Would love to see this land. Happy to share repro logs if useful.