The AI tracker: From Indian language models to hiring headaches

Indian startup Sarvam AI has launched Sarvam Audio, an audio-first language model designed to understand and transcribe real-world speech across India’s multilingual landscape. Built on the company’s 3-billion-parametre Sarvam 3B model, the system supports transcription in 22 Indian languages, including Hindi, Tamil, Telugu, Malayalam, Bengali and Indian English. Unlike global rivals such as ElevenLabs, which focus on expressive voice generation, Sarvam Audio prioritises speech understanding and transcription. The company claims the model outperforms GPT-4o-Transcribe and Gemini-3-Flash on the IndicVoices dataset, achieving lower word error rates across unnormalised, normalised and code-mixed transcription styles.