Sarvam Prism

Released
2026-11-03
Last refreshed
2026-11-09
Status
Researched 2d ago
Open sourceCommercial use: permittedRAGLong contextClassificationJSON / Tool use
Sarvam.ai releases · 2 in the last 12 monthsChangelog →
Specifications
Family
Sarvam
Released
2026-11-03
Context
128K
Max output
0
Parameters
105B (10B active)
Architecture
Mixture of Experts
Specialization
classification
Openness
Open source
License
Apache 2.0OSI-approvedPermitted
Weights
Available
Code
Unknown
Training
Fine-tuned
Created by
Sarvam.ai

Multilingual Voice AI for India

Bengaluru, India
Founded 2023
Website
Pricing
Output / 1M
₹0
Input / 1M
₹4

Sarvam AI

Providers(1)
Sarvam AI
Links
Website HuggingFace

About

Sarvam Prism is Sarvam AI's open-weight decision model, released November 3, 2026. Trained on the Sarvam 105B mixture-of-experts base (105B total, 10B active), it answers yes/no, pick-one and ordered-scale questions about an input with a probability for every option instead of generating text, for routing, triage, moderation and judging agent steps. Weights are on Hugging Face at sarvamai/prism under Apache 2.0 and run in FP8 on one GPU with about 120 GB of memory. It is also served on Sarvam's Decisions API as sarvam-prism.

Providers(1)

Compare API pricing across 1 tracked provider route for input and output tokens, batch, and cached reads when available.

ProviderInput / 1MOutput / 1MRoute
Sarvam AI₹4₹0
ServerlessPartial

Capabilities

Structured Outputs

Benchmark peer barsfor RAG

No task-mapped benchmark peers are available for this model yet.

Migration checks

No linked migration route is available for this model yet.

API versions

sarvam-prismsarvamai/prism