Sarvam AI brings 24B-parameter LLM for Indian languages, reasoning
The Bengaluru startup noted that Sarvam-M sets a new benchmark for models of its size in Indian languages, as well as in math and programming tasks.
Artificial intelligence (AI) startup Sarvam AI has introduced a 24-billion-parameter large language model (LLM) designed for Indian languages and to handle reasoning tasks such as math and programming.
The model, Sarvam-M (where M stands for Mistral), is an open-weights hybrid model based on Mistral Small, which is a compact yet powerful open-source language model.
It has been enhanced by training it on carefully selected examples to improve accuracy (a process known as supervised fine-tuning). It also learns from feedback tied to clear, measurable goals, such as solving a math problem correctly, which helps it make better decisions (called reinforcement learning with verifiable rewards).
Additionally, the model has been optimised to respond more efficiently and accurately when generating answers, especially during real-time use.
The Bengaluru startup noted that Sarvam-M sets a new benchmark for models of its size in Indian languages and math and programming tasks.
According to a blog post, Sarvam-M shows major improvements over the base model, with average performance gains of 20% on Indian language benchmarks, 21.6% on math tasks, and 17.6% on programming tests.
The model performs even better on tasks that combine Indian languages and math. For example, it achieved an 86% improvement on a romanised Indian language version of the GSM-8K benchmark, it added.
“In most benchmarks, our advanced Sarvam-M outperforms Llama-4 Scout, is comparable to larger dense models like Llama-3.3 70B, and models like Gemma 3 27B, which are pre-trained on significantly more tokens,” the blog post noted.
Sarvam AI added that one area where the model still needs improvement is in English knowledge-based benchmarks like MMLU, where Sarvam-M performs about 1 percentage point lower than the baseline model.
According to the AI firm, Sarvam-M is built for versatility and designed to support a wide range of applications, including conversational agents, translation, and educational tools.
It is available for download on Hugging Face, can be tested on Sarvam AI’s playground, and accessed through its APIs for development.
The startup plans to release models regularly as part of its effort to build a sovereign AI ecosystem in India. This model is the first in that series of contributions.
In late April, the Indian government selected Sarvam to build the country’s sovereign LLM as part of the IndiaAI Mission, a national effort to strengthen domestic capabilities in emerging technologies.
Edited by Suman Singh