Today we’re releasing Sarvam Prism, a System One model built on...

1

Today we’re releasing Sarvam Prism, a System One model built on Sarvam 105B for fast, accurate decisions across enterprise workflows.

It scores 71.8 with a median response time of 44 milliseconds, the highest average score among decision models that respond in under 100 milliseconds.

2

Across three decision indices, Prism advances the Pareto frontier of accuracy and speed among the 25 models we evaluated.

3

For an enterprise to automate a decision, it has to trust the model’s confidence. Across every level we tested in English and Indian languages, Prism made fewer errors than its scores predicted.

4

We also tested whether changing the order of the options changed its decisions. Across 2,000 questions, Prism gave the same answer every time we shuffled the options, with no extra computation or loss in accuracy.

5

Prism has found its way into many AI applications, from speech to speech harness and trace evaluation to prompt guardrails and router for token factory. We’re sharing benchmarks of how Prism performs on trace evaluation versus other models.

6

Every enterprise decides differently. We’re releasing Prism with open weights so enterprises can fine-tune it on their own outcomes using simple training recipes and host it efficiently within their own infrastructure. We also support training and deployment on Sarvam’s infrastructure.

Read the technical blog: links.sarvam.io/prism-research

Prism’s weights are on Hugging Face: huggingface.co/sarvamai/prism

It's available as an API: links.sarvam.io/prism-api

Today we’re releasing Sarvam Prism, a System One model built on...

Actions

What You Can Do

Actions

What You Can Do

More in AI & Machine Learning