Model family timeline
Last updated Mar 6, 2026
Sarvam model releases
A source-backed timeline for the Sarvam model family, collecting release dates, labs, access details, context windows, and major lifecycle changes.
3
Models
1
Labs
3
Open
3
Recent
Most recent in this set
3 models
Sarvam-105B
AvailableSarvam AIOpen source
Apache-licensed Indian-context MoE from Sarvam AI, optimized for reasoning, coding, agentic tasks, and 22 Indian languages.
Type
MoE
Params
105B
Context
128K
Top eval
—
Released
Mar 6, 2026
4mo ago
MoE105B128K ctxMar 6, 2026
Sarvam-M
AvailableSarvam AIOpen weights
Sarvam's medium-scale open model for multilingual Indian-language chat, reasoning, and translation tasks.
Type
Dense
Params
Undisc.
Context
—
Top eval
—
Released
May 21, 2025
1.1y ago
DenseUndisc.— ctxMay 21, 2025
Sarvam-1
AvailableSarvam AIOpen weights
Sarvam's 2B open model trained for ten major Indian languages.
Type
Dense
Params
2B
Context
—
Top eval
—
Released
Oct 22, 2024
1.7y ago
Dense2B— ctxOct 22, 2024