Use · Sarvam-30B, Sarvam-105B and Sarvam-105B-1711

Is it good enough?

Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input D ยท 44.4/100

This page is a projection of the one entry record, the Reliability factor that Use covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.

Which domain expands which factor
  • AssessUse & modify + Transparency
  • ImplementData control + Reliability
  • UseReliability
  • SupportTransparency

Capabilities & modalities

Text-only reasoning models for chat, mathematics, code, tool calling and agentic tasks. The 105B targets agentic and reasoning work. The 30B targets low-latency conversation and voice-agent tool calls. The 1711 build adds post-training for tool calling, instruction following, agentic tasks and Indian languages (ev-card-105b, ev-card-30b, ev-card-1711).

All scores are the publisher's own. Artificial Analysis ranks both March builds below its listed peers (ev-aa).

Context window & long-context behaviour

The 105B card states 128K tokens through YaRN scaling. The 30B card states 65,536 tokens as its evaluation length, and its config allows 131,072. The 1711 examples use 131,072, and the config allows 163,840 (ev-card-105b, ev-card-30b, ev-card-1711, ev-hub).

No long-context retrieval benchmark is published. Test your own long inputs.

Prompt format & chat template

Each repo bundles chat_template.jinja. Call apply_chat_template with enable_thinking=True or False. The 1711 cards recommend temperature 1.0 with thinking and 0.6 without, and thinking is on by default (ev-card-105b, ev-card-1711).

Language coverage

The cards tag 22 language codes plus English. The tags include Tibetan (bo) and omit Bodo (brx), so they are not the 22 scheduled languages exactly (ev-card-105b).

The blog says pre-training emphasised the 10 most-spoken Indian languages and the tokenizer covers all 22 scheduled languages (ev-blog). Sarvam's API documentation lists 10 Indian languages plus English for the 105B and the 30B (ev-api-docs). The 10 are Hindi, Bengali, Tamil, Telugu, Gujarati, Kannada, Malayalam, Marathi, Punjabi and Odia.

The only 22-language evidence is a pairwise LLM-judge comparison over the public indivibe prompts, with average win rates reported and per-language results shown as charts (ev-blog, ev-indivibe). The 1711 Indic Suite is internal and has no language list. Treat the other 12 languages as unverified.

Function / tool calling

The blog describes native tool calling. The 1711 build is further post-trained for it and the cards use vLLM's glm45 tool parser. Publisher scores: tau2-Bench 85.5 and BFCL v4 62.4 for 1711, tau2-Bench 68.3 for the 105B and 45.7 for the 30B (ev-blog, ev-card-1711, ev-card-105b, ev-card-30b).

Test the parser on your own serving stack and sandbox tool execution.

Structured / JSON-constrained output

The hosted API documentation lists JSON and json_schema outputs. No card or cookbook documents a constrained-output mode for the downloadable weights. That is a gap.

How this scores

The ownership factor this domain covers, drawn from the one entry record.

3

ReliabilityIs it reliable and good enough for the job?

Weak

Safety scores 1: no safety evaluation, no guard model and silent 1711 cards. Operational scores 3 (ev-blog, ev-card-1711, ev-hub).

How this scores (AOI sub-dimensions)
Operational3/5how practical it is to run, serve and maintain in productionThe models run on Transformers with trust_remote_code, SGLang and vLLM.
Safety1/5whether misuse risks are evaluated and guardrails are providedThe blog says the fine-tuning data covered India-specific risks and red-team prompts.
What this means for adoptionYou can run and fine-tune Apache-2.0-declared weights, but you get no data, code, recipes or safety case, and the licence is a card declaration with no LICENSE file. Sarvam's own Terms and EULA restrict derivative works and training without mentioning the weights. Ownership rises if Sarvam adds a LICENSE file, clarifies those documents and publishes a safety evaluation.

Sources

The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.

Model cardread2026-11-19
The 105B card declares Apache-2.0 and lists 22 language codes plus English.
Model cardread2026-11-19
The 30B card declares Apache-2.0 and the same language codes.
Model cardread2026-11-19
The 105B-1711 cards declare Apache-2.0 and say the model shares the Sarvam-105B base and is further post-trained for tool calling, instruction following, agentic tasks and Indian languages.
Licenceread2026-11-19
All eight weights repos declare Apache-2.0 in the card front matter and none holds a LICENSE file (raw LICENSE returns HTTP 404).
Documentationread2026-11-19
The sarvamai Hugging Face organisation is not verified and lists 17 ungated model repos.
Vendor announcementread2026-11-19
The launch blog of 6 March 2026 says both models were trained from scratch on in-house data, 16T tokens for the 30B and 12T for the 105B, with the 10 most-spoken Indian languages emphasised in pre-training (publisher claims).
Documentationread2026-11-19
The indivibe dataset holds 4,840 evaluation prompts in the 22 scheduled languages, including Bodo, in native and romanised script, under CC-BY-4.0.
Documentationread2026-11-19
Sarvam's API documentation lists 10 Indian languages plus English for Sarvam-105B and for the deprecated Sarvam-30B.
Documentationread2026-11-19
The platform documentation says workspace data is retained indefinitely until an owner sets a period.
Terms of serviceread2026-11-19
Terms of Service v2.0, effective 29 July 2026, clause 10.5 bars using the Offerings or their output to train or fine-tune AI systems and bars discovering model weights.
Privacy Policyread2026-11-19
The Privacy Policy of 29 July 2026 headlines model training as opt-in but says Sarvam trains on inputs and outputs unless the user opts out.
Terms of serviceread2026-11-19
The EULA of 22 May 2026 defines Software to include 'proprietary language models' and bars derivative works.
Securityread2026-11-19
The Trust Center claims ISO 27001, SOC 2 Type II, India-only residency and no use of one customer's data to train models for other customers.
Third-party analysisread2026-11-19
Artificial Analysis scored Sarvam 105B at 18 and Sarvam 30B at 12 on its Intelligence Index in April 2026.
Documentationread2026-11-19
Official FP8 builds of both models and Q4_K_M GGUF builds (19.6 GB for the 30B, 64.2 GB for the 105B) exist.
Documentationread2026-11-19
Community threads ask for the training data and for newer checkpoints and have no Sarvam reply.
Vendor announcementread2026-11-19
The Token Factory post of 16 November 2026 says Sarvam serves the 105B in MXFP8 with DSpark speculative decoding from GPU clusters in India.
Technical_reportunverified2026-11-19
No technical report, paper or data statement for these models was found.
Documentationunverified2026-11-19
The AIKosh page for the 105B, linked from the blog as a second download source, is a JavaScript app and returns no text.