Sarvam launches new decision-making model 'Prism'

India’s sovereign LLM company Sarvam has launched a new model, ‘Prism’, built on a System One decision-making architecture. It is designed for quick decision-making in situations where the context is already established and decisions involve choosing from available options rather than generating text-based answers.

The company in its blogpost said that Prism fared much better compared to 25 other decision models that it tested on the Pareto frontier of accuracy and speed. "The model scores 71.8 while responding in just 44 ms on B200 GPU, the highest-scoring model under 100 ms across the 25 models we tested, averaged over three indices," the company said in its post.

The model when applied where AI looks at the information provided and returns a structured result, such as selecting an option, giving a score, or estimating a probability, Sarvam claimed in Indian languages, of the error expected from Prism being 5.4% errors, it fared much better at 4.4% while other models made 2.7 to 4.8 times as many errors as their confidence scores predict.

"For English answers at 90% confidence or higher, Prism predicts 3.3% errors and makes 2.2%. Clef is close, predicting 2.8% and making 2.3%. Perplexity Decider, Microsoft-Decision-1 and Jev 1.14 underestimate their errors by factors of 1.5, 1.8 and 2.7, respectively," Sarvam noted in its charts.

The company also announced releasing the open weights of the model for enterprises to train this calibrated base on their own data and run it on their own infrastructure. Sarvam also said that it has tested Prism on real-world use-cases like speech-to-speech harness, trace evaluations, prompt injections guardrails, intent routing for customer support, model routing, and document intelligence, and would release the benchmarks in the coming days.

According to a report by data intelligence platform Tracxn, the Indic AI ecosystem raised $657 million between 2021 and October 2026, with this year marking much of the capital coming with Sarvam’s 2 rounds of $255 million and $90 million, accounting for $346 million, about 87% of the 2026 funding. In July this year, HCL Tech, one of Sarvam’s investors, announced spending ₹14,257 crores to set up the AI Data Center in partnership with Sarvam in Odisha, including financial assistance from the state government.