Indian artificial intelligence (AI) startup Sarvam AI, the first company selected by the government to build a sovereign foundational model under the India AI Mission, recently launched its multilingual open-source large language model (LLM) called Sarvam M. While Sarvam AI claims the LLM has achieved strong benchmarks across mathematics, programming, and a range of Indian languages, its launch has elicited a tepid response, raising questions about the current state of India's AI ecosystem, particularly given the model's low initial download numbers.

Sarvam M is a 24-billion-parameter model, developed on top of the 'Small' model from French AI company Mistral. According to the startup, Sarvam M is "built for versatility" and is designed to support a diverse array of applications, including conversational agents, translation services, and educational tools. The model provides support for 10 Indian languages, such as Hindi, Bengali, Gujarati, Kannada, and Malayalam. In a post on X, Sarvam AI stated, "This is the first in a series of contributions as we help build out the Sovereign AI Ecosystem in India."

The model comes with several key features. It incorporates a Hybrid thinking mode, which enables support for complex logical reasoning, the ability to solve mathematical problems, and the capacity to handle coding tasks; it also features a 'non-think' mode for general-purpose conversations. For Indic skills, Sarvam M has been post-trained in various Indian languages alongside English to better cater to local needs. Regarding its Reasoning capabilities, the company asserts that Sarvam M surpasses similarly sized models on benchmarks for coding and mathematics.

Despite these features, the model's initial traction was modest. In the first two days following its launch, Sarvam M garnered just over 300 downloads. By May 27, this number had increased to 1,200 downloads on the Hugging Face platform. The model can be tested on Sarvam AI's dedicated playground and accessed through its APIs. This relatively low uptake sparked a debate on social media platforms concerning the standing of India's AI development efforts when compared to global competitors like OpenAI and China's DeepSeek.

The discussion was intensified by criticism from figures such as Deedy Das, a venture capitalist at Menlo Ventures. In a critical post on X, Das highlighted the muted launch of what he termed "India's biggest AI startup $1B Sarvam," noting only 23 downloads in the initial two days. He contrasted this with a Korean open-source model developed by two college students, which reportedly achieved around 200,000 downloads in the previous month. Das later clarified in another post that he held nothing against Sarvam or India's attempts to build AI capabilities but was "disappointed at their direction."

In response to the criticism, many users came to Sarvam AI's defense. One user pointed out the irony: "Before Sarvam M came out, people were complaining about the lack of IndicLLMs. After Sarvam M came out, people are still complaining." Another commenter emphasized the need to offer encouragement to entrepreneurs rather than shame them, stating, "Teams in India are trying and they are working to build the muscle and culture of what good or amazing looks like." Sridhar Vembu, CEO of Zoho, also defended Sarvam, arguing that there was "no such thing as an 'instant hit'." He added, "Even when we were the first mover in a new market and we had done a lot of technical work, we only got slow traction. Instant success is neither necessary nor sufficient to succeed long term," and encouraged the Sarvam team to "fight the good fight."

Sarvam AI, based in Bengaluru, holds a significant position in India's AI strategy as the first startup chosen to develop an indigenous foundational model under the Rs 10,000 crore IndiaAI Mission. As part of this initiative, Sarvam AI was granted access to 4,096 Nvidia H100 graphics processing units (GPUs) for six months from the mission's common compute cluster to train its model. According to data from Tracxn, the company has raised a total of $53.6 million in funding and is currently valued at $111 million.