Description

Summary

• Previously, token-limit diagnostics in the Gemini and OpenAI providers only ran when the returned text was empty (for example, when reasoning tokens consumed the entire output budget). When a model stopped at the output token limit (finish_reason=MAX_TOKENS / numeric 2 in Gemini, or finish_reason=length in OpenAI) after emitting non-empty partial text, the truncated payload reached Resolver silently and surfaced only as ResolverParsingError: Failed to parse JSON content: Unterminated string ... (or was dropped without explanation when suppress_parse_errors=True).
• Log an actionable logging.warning naming the finish_reason and pointing to max_output_tokens and max_char_buffer whenever a non-empty response stops at the output token limit across:
  • GeminiLanguageModel._response_to_scored_output (langextract/providers/gemini.py)
  • Gemini Batch JSONL / in-memory parsing (_is_max_tokens_finish and _parse_batch_line in langextract/providers/gemini_batch.py, supporting string "MAX_TOKENS", numeric 2, and FinishReason.MAX_TOKENS enum values while rejecting bool)
  • OpenAILanguageModel._response_to_scored_output (langextract/providers/openai.py)
  • OpenAI Batch JSONL parsing (_extract_text_from_response_body and _collect_outputs_from_jsonl in langextract/providers/openai_batch.py, including the batch item custom_id)
• Preserve existing return values, empty-response InferenceRuntimeError behavior, downstream Resolver parsing, and public signatures unchanged.

Fixes #287

Bug fix

How Has This Been Tested?

Tests

• Focused unit tests across all four provider paths (266 passed under both google-genai==2.29.0 and google-genai==3.0.0):
  • pytest tests/gemini_retry_test.py tests/test_gemini_batch_api.py tests/inference_test.py tests/openai_batch_test.py -q
• Formatting and lint checks (10.00/10 on both source and test modules):
  • isort --check-only and pyink --check --config pyproject.toml on all modified files
  • pylint --rcfile=.pylintrc langextract/providers/gemini.py langextract/providers/gemini_batch.py langextract/providers/openai.py langextract/providers/openai_batch.py
  • pylint --rcfile=tests/.pylintrc tests/gemini_retry_test.py tests/test_gemini_batch_api.py tests/inference_test.py tests/openai_batch_test.py
• PR CI on 9de5b6bbcd722c23a3fea7dfd316fb69bacdce50: https://github.com/google/langextract/actions/runs/42734188186 (format-check, Python 3.10/3.11/3.12 unit + lint suites, ollama-integration-test, and plugin-integration-test passed).
• Secure fork live-API dispatch on 9de5b6bbcd722c23a3fea7dfd316fb69bacdce50: https://github.com/google/langextract/actions/runs/42735393166 (12 passed live Gemini and OpenAI tests; 4 skipped opt-in Vertex AI Batch and live OpenAI Batch tests).

Checklist

• I have read and acknowledged Google's Open Source Code of conduct.
• I have read the Contributing page, and I either signed the Google Individual CLA or am covered by my company's Corporate CLA.
• I have discussed my proposed solution with code owners in the linked issue(s) and we have agreed upon the general approach.
• I have made any needed documentation changes, or noted in the linked issue(s) that documentation elsewhere needs updating.
• I have added tests, or I have ensured existing tests cover the changes
• I have followed Google's Python Style Guide and ran pylint over the affected code.