Archives

This study benchmarks quantized variants of LLMs (Qwen2.5, DeepSeek, Mistral, and LLaMA 3.3) across diverse tasks to assess the trade-offs in accuracy and robustness.

A practical and neutral benchmark of seven leading STT models, including open-source and proprietary APIs, based on real-world transcription challenges and diverse audio.