Sarvam to Host "Epoch" AI Event on July 30; Model Updates and Product Launches

Sarvam AI is hosting Epoch later this month. The Indian AI startup describes Epoch as a cutting-edge AI conference and has offered hints about what to expect. From new AI models to product launches, here is everything you need to know about Sarvam Epoch.

Sarvam shared a preview of Epoch on X. In the video, creator Caleb Friesen explained the term's significance: "Epoch is a word taken from the world of machine learning. During training, an 'epoch' consists of a complete pass through the data. In this process, the model processes all the information, updates itself, and becomes more capable."

Notably, Epoch is Sarvam's first major event since its latest funding round. The company recently raised $234 million at a $1.5 billion valuation—what it termed the largest Series B round in the history of Indian startups. The round was led by HCLTech, which contributed $150 million as a strategic investor.

Sarvam Epoch: AI Model Updates and More

Epoch will feature two versions: the "Builder Edition" and the "Enterprise Edition."

Sarvam has teased that it will share news regarding AI model updates and launch new products during the Builder Edition. The Indian startup is also likely to present its latest research and showcase the technical work happening behind the scenes.

The AI startup currently has several models, such as Sarvam-105B and Sarvam-30B. The company's Sarvam Vision model previously outperformed systems like ChatGPT and Gemini in reading documents in Indian languages. In May, Sarvam cut the price of its Vision AI model by 67% as developers and partners used the platform to digitize over 35 million pages.

The startup also plans to host hands-on workshops, in-depth technical sessions, and live demonstrations. A famous publication says Epoch will focus on how organisations can adopt AI tools into their daily workflows. Sarvam Epoch will take place in Bengaluru on July 30 and 31, 2026.

Sarvam AI invests millions in compute capacity

Ahead of the event, Sarvam co-founder and CEO Pratyush Kumar stated that the Bengaluru-based startup had invested nearly $20 million in compute capacity to develop three models, including its flagship model, Sarvam-105B.

Kumar explained that while the total cost ran into the millions, it was still lower than that of companies like China’s DeepSeek. "The compute cost was undoubtedly significantly lower than what DeepSeek required. In fact, the total compute expenditure was around $20 million, or even slightly less. And we trained three models," he said on an NPCI podcast.

The Sarvam CEO clarified that this figure did not represent the cost of a single training run, but rather the cumulative expenditure throughout the entire model launch process. "It’s not the cost of a single execution. It involves a lot of exploratory work and extensive data creation that gets integrated in layers. But, in short, we aren't talking about billions of dollars; it’s a model with a cost in the single-digit millions range," he explained.

Much of Sarvam's training was conducted using compute resources for which the company did not have to pay directly. It was the first company selected under India’s sovereign AI mission, an initiative in which the government agreed to provide access to 4,096 Nvidia H100 GPUs for six months in exchange for an equity stake in the startup.

The company has also highlighted the growing demand for its products. Sarvam states that its platforms currently handle over 2 million daily interactions, process nearly 10 million API requests per day, and transcribe more than 500,000 hours of audio each month.

Sarvam’s multilingual voice agents, used in a government program, gathered information from 17 million farmers, while a national insurance campaign reached 45 million policyholders.