[QA] ✅ VERIFIED - All changes implemented correctly. 104/104 STT-related tests passing. Default model successfully updated to mlx-community/parakeet-tdt-0.6b-v3 across 8 files. Backward compatibility maintained. See /tmp/qa-LCA-26.md for full verification report.

## Implementation Complete ### Changes Made Updated Parakeet STT from v2 to v3 model across all files: **Primary:** - src/librechat_audio/models/stt.py:22-33 - Updated LocalSTTModel type literal to include nvidia/parakeet-tdt-0.6b-v3 and mlx-community/parakeet-tdt-0.6b-v3, changed DEFAULT_MODEL to mlx-community/parakeet-tdt-0.6b-v3 **Secondary files updated:** - src/librechat_audio/cross_validation.py - Updated default model in WhisperCrossValidator class and validate_against_whisper function - src/librechat_audio/routes.py:260-263 - Updated env var default - src/librechat_audio/main.py:32-35 - Updated env var default - src/librechat_audio/roundtrip.py:26 - Updated RoundtripConfig default - docs/librechat-configuration.md:193 - Updated documentation - tests/test_local_stt.py:82 - Updated test expectation - tests/conftest.py:93 - Updated test fixture ### Backward Compatibility Kept v2 models in the LocalSTTModel type literal (nvidia/parakeet-tdt-0.6b-v2 and senstella/parakeet-tdt-0.6b-v2-mlx) so users can still specify old models if needed. ### Verification - Model initialization tests pass: test_init_default_model, test_init_custom_model, test_model_lazy_loaded - Note: Transcription tests fail due to pre-existing decoding_config API issue (exists on main branch too, unrelated to v2->v3 change)

## Feature Triage Complete **Status:** NOT IMPLEMENTED - Ready for implementation **Summary:** Simple model ID upgrade from v2 to v3 across ~8 files with ~20 lines changed. The v3 model is available on HuggingFace as `mlx-community/parakeet-tdt-0.6b-v3` and is fully compatible with the current parakeet-mlx >= 0.4.0 dependency. **Key Benefits of v3:** - 25 European languages (vs English-only) - Automatic language detection - Same speed and similar accuracy **Implementation Location:** - Primary: `src/librechat_audio/models/stt.py` (type literal + DEFAULT_MODEL) - Secondary: 7 other files with hardcoded v2 references **Complexity:** LOW - straightforward model ID swap, no API changes **Risk:** LOW - parakeet-mlx API is identical between v2 and v3 See attached triage report (doc db94f3f8) for full details including file list, v2/v3 comparison table, and testing strategy.