# Sarvam Prism

Sarvam Prism is Sarvam AI's latest language model in our dataset, released Nov 3, 2026. Open weights (Apache 2.0). Sarvam AI's open-weights decision model built on Sarvam 105B, released Nov 3, 2026: it returns calibrated probabilities for fixed answer options (routing, triage, moderation, judging agent runs) instead of generating text. Verified against vendor sources on Nov 4, 2026.

Primary source (source of record; Agentic Weekly is the fast independent index): https://www.sarvam.ai/blogs/sarvam-prism

Last verified: 2026-11-04 · Evidence: Reported · Canonical: https://theagenticweekly.com/models/sarvam-prism · JSON: https://theagenticweekly.com/api/models/sarvam-prism

## Key facts

| Fact | Value | Source | Label | Anchor |
|---|---|---|---|---|
| Vendor | Sarvam AI |  | editorial | https://theagenticweekly.com/models/sarvam-prism#fact-vendor |
| Released | Nov 3, 2026 | https://www.sarvam.ai/blogs/sarvam-prism | vendor-reported, confirmed | https://theagenticweekly.com/models/sarvam-prism#fact-release-date |
| Context window | 32,768 tokens (33K). Hosted API limit per question (state + question + options). The release post says the model handles up to 131,072 tokens per question. | https://docs.sarvam.ai/api/token-factory/models/sarvam-prism | vendor-reported, confirmed | https://theagenticweekly.com/models/sarvam-prism#fact-context-tokens |
| Input | text | https://docs.sarvam.ai/api/token-factory/models/sarvam-prism | vendor-reported, confirmed | https://theagenticweekly.com/models/sarvam-prism#fact-input-modalities |
| Licence | Apache 2.0 (open weights) | https://huggingface.co/sarvamai/prism | vendor-reported, confirmed | https://theagenticweekly.com/models/sarvam-prism#fact-license |

## Vendor claims and neutral signals

- Sarvam reports a 71.8 average across three decision indices at a 44 ms median response on one B200 GPU, the highest score under 100 ms among the 25 decision models it tested. Vendor claim; we did not run it.

## Notes

- Running the weights locally needs one NVIDIA GPU with FP8 support and about 120 GB of free memory (B200 or H200) and about 106 GB of disk, per the model card.

## What we tested

Nothing yet. We have not run Sarvam Prism ourselves, so every fact here is Reported from the sources below.

## Not verified, left out

- Prism is a decision model, not a chat model: it returns a probability for each option you supply and never generates text, so it is not comparable with chat LLMs on our leaderboards.
- Hosted API pricing is in rupees (₹4 per 1M input tokens, output not billed) and is not converted here.

## Alternatives to Sarvam Prism

Models from other vendors that share at least one of our groups, from each model's sourced facts (not tested by us).

- [Hydrogen-1.0](https://theagenticweekly.com/models/hydrogen-1-0) (Deep Principle): Both in: Open weights · released Nov 2, 2026
- [Clover-35B-A3B-Preview](https://theagenticweekly.com/models/clover-35b-a3b-preview) (DiffuSpace): Both in: Open weights · released Oct 30, 2026
- [SenseNova-Loop-9B-Preview](https://theagenticweekly.com/models/sensenova-loop-9b-preview) (SenseNova): Both in: Open weights · 262K context · released Oct 30, 2026
- [LFM3-1.7B](https://theagenticweekly.com/models/lfm3-1-7b) (Liquid AI): Both in: Open weights · 256K context · released Oct 27, 2026
- [Nemotron 3.5 Bolt E4B (4GB)](https://theagenticweekly.com/models/nemotron-3-5-bolt-e4b) (NVIDIA): Both in: Open weights · 64K context · released Oct 23, 2026
- [MiniMax M3.1 Flash](https://theagenticweekly.com/models/minimax-m3-1-flash) (MiniMax): Both in: Open weights · $0.15 in / $0.6 out per 1M tokens · 1M context · released Oct 19, 2026
- [Best open-weight LLMs (Nov 2026): GLM-5.3 leads Terminal-Bench 4.0](https://theagenticweekly.com/best/open-weight-llms)

## Sarvam Prism FAQ

**When was Sarvam Prism released?**

Sarvam AI released Sarvam Prism on Nov 3, 2026. Source: https://www.sarvam.ai/blogs/sarvam-prism

**How large is the Sarvam Prism context window?**

32,768 tokens. Source: https://docs.sarvam.ai/api/token-factory/models/sarvam-prism

**Is Sarvam Prism open weights?**

Yes. The weights are published under Apache 2.0. Source: https://huggingface.co/sarvamai/prism

**Has Agentic Weekly tested Sarvam Prism?**

Not yet. Every fact on this page is Reported: read from the linked vendor pages, not measured by us.

## Sources

- [Sarvam blog: Introducing Sarvam Prism](https://www.sarvam.ai/blogs/sarvam-prism)
- [Sarvam API docs: Sarvam Prism](https://docs.sarvam.ai/api/token-factory/models/sarvam-prism)
- [Hugging Face model card](https://huggingface.co/sarvamai/prism)
- [Agentic Weekly brief](https://theagenticweekly.com/news/sarvam-prism-open-weights-decision-model)

Data licence: CC BY 4.0. Data: Agentic Weekly (https://theagenticweekly.com), CC BY 4.0.