Amid a wave of criticism targeting Sarvam AI´s newly launched large language model, Sarvam-M, Zoho founder and chief scientist Sridhar Vembu offered a strong public defense of the homegrown Artificial Intelligence startup. Vembu, leveraging his own experiences in building global SaaS products from India, argued that instant market success is uncommon, even for technically accomplished and first-mover products. He emphasized that slow initial traction does not determine a technology´s long-term value and encouraged Sarvam AI to persist in its efforts.
The controversy began when Deedy Das, a VC at Menlo Ventures, disparaged Sarvam-M’s metrics on Hugging Face, where it had recorded only 23 downloads two days post-launch. Das contrasted these numbers with a South Korean open-source project’s more rapid adoption, calling Sarvam-M’s debut ´embarrassing´ and suggesting the Indian Artificial Intelligence scene was focusing on trendy endeavors rather than substantive problem-solving. Das later clarified that his frustrations centered around unmet expectations, not opposition to Sarvam AI’s ambitions or the broader direction of India´s Artificial Intelligence efforts.
Sarvam-M, announced as India’s flagship sovereign Large Language Model (LLM) under the government-backed IndiaAI Mission, is a 24-billion parameter, multilingual, hybrid-reasoning model based on Mistral Small, a compact and efficient foundation LLM from France´s Mistral AI. The company claims superior performance across Indian languages, mathematics, and programming tasks for its size, targeting applications such as conversational agents, translation, and educational tools. Sarvam-M is open-sourced on Hugging Face, with over 1,200 downloads reported as of this article, and developer APIs available for wider use. Sarvam AI’s co-founders, Pratyush Kumar and Vivek Raghavan, maintain a positive outlook, inviting feedback to further advance their efforts to build robust, indigenous foundational models for India. The startup, founded in July 2023, has been selected for dedicated government compute resources to further scale Generative Artificial Intelligence for the Indian market.
