Sarvam

Sarvam 105B

CapableOpen WeightsReasoningApache 2.0
Price
Free
Self-host
Intelligence
12
/100
Context
128K
tokens
Parameters
105B MoE (10.3B active)

One9Founders Verdictgood

India's flagship open-weight MoE (105B total, 10.3B active), trained from scratch in India under the IndiaAI mission. Apache 2.0. Strong Indic-language coverage (22 official languages). AA ~12 — not a global frontier model, but the sovereign default for Indic + DPDP-sensitive work.

The Indian open-weight model to actually evaluate.

Best For

  • Indic languages
  • India-sovereign / DPDP stacks
  • government and public-sector apps

Not Great For

  • frontier English coding agents
  • multimodal

Startup Recommendation

Shortlist for Hindi/Indic products and any stack that must stay in India.

Capabilities

-Function Calling
-Structured Output
-Vision
-Web Search
-Code Execution
-MCP
-Audio Input
-Video Input
Reasoning
Streaming
-Batch API
-Computer Use

Security & Compliance

SOC 2 Type II
GDPR
HIPAA
Enterprise SSO
Audit Logs

Training data policy: N/A — open weights. Self-host for full control.

DPDP Act (India): Trained in India. Self-host or use India-region API for DPDP control.

Data residency: India, Self-hosted

India Availability

Self-host in India