Sarvam AI's open-weight System One decision model that scores fixed options in a single pass and returns calibrated probabilities for enterprise workflows.
Sarvam Prism launch
Overview
What is Sarvam Prism?
A fast, open-weight decision model that answers in 44 milliseconds.
Sarvam Prism is a System One decision model from Sarvam AI, released on November 3, 2026. Built as a LoRA fine-tune of the Sarvam 105B model, it answers typed questions by scoring fixed options in a single forward pass instead of generating text.
- 44 ms median response time on one NVIDIA B200 GPU, with a 71.8 average score across three decision indices, according to Sarvam
- Open weights on Hugging Face under the Apache 2.0 licence; runs in FP8 on one GPU with at least 120 GB of memory
- Also available as a hosted API on Sarvam Indus, with documentation on the Sarvam docs site
Prism is built for enterprise workflows where a fast, calibrated yes/no, choice, or score decision matters more than generated prose.
Platforms and languages
Sarvam Prism Availability
Platforms
Languages
Capabilities
Sarvam Prism Key Features
Single-pass decisions
Scores fixed options in one forward pass instead of generating text token by token, returning a probability for each option.
44 millisecond responses
Sarvam reports a 44 ms median response time on a single NVIDIA B200 GPU in FP8, with an average score of 71.8 across three decision indices.
Calibrated confidence
Confidence scores are designed so an enterprise can automate only above a threshold; Sarvam reports errors stay within the stated confidence bounds.
Open weights, Apache 2.0
Weights are published on Hugging Face under Apache 2.0, so enterprises can fine-tune Prism on their own data and run it on their own infrastructure.
Long-context evaluation
Handles up to 131,072 tokens per question, so long agent traces can be evaluated in one pass.
Best for
Who uses Sarvam Prism?
Enterprise teams automating routine decisions
Route support tickets, evaluate agent runs, and score documents with calibrated confidence
Plans and access
Sarvam Prism Pricing
Free
Open weights under Apache 2.0; hosted API pricing not published.
Common questions
Sarvam Prism FAQs
What is Sarvam Prism?
Prism is Sarvam AI's System One decision model. It scores fixed options in a single pass and returns calibrated probabilities, instead of generating text.
How fast is Prism?
Sarvam reports a 44 millisecond median response time on one NVIDIA B200 GPU, with an average score of 71.8 across three decision benchmarks it ran.
Is Prism open source?
Its weights are released under the Apache 2.0 licence on Hugging Face, so enterprises can fine-tune and self-host it.
What hardware does Prism need?
It runs in FP8 on a single GPU with at least 120 GB of memory, according to Sarvam.
How is Prism different from a chatbot model?
It never generates text. You send a state, a question, and fixed options, and it returns a probability for each option, which suits routing, evaluation, and guardrail decisions.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.
