Sarvam AI’s Open Source Bet Faces Early Adoption Hurdles

Sarvam AI has open-sourced two large reasoning models, Sarvam 30B and Sarvam 105B, positioning them as India-built foundation models optimised for Indic languages and agentic workloads.

While the move signals a push to seed an Indian open-source AI ecosystem, developers say the release currently faces friction due to missing tooling support and deployment formats.

The bigger strategic question remains: can Sarvam build a developer ecosystem before global players release stronger Indic-focused models?

Added to Saved Stories in Login VIEW SAVED STORIES .inc42-toggle-item-popup { display: none; position: relative; } .toggle-item-close { text-align: end; padding: 8px 12px 0px 10px; position: absolute; right: 0; cursor: pointer; } .toggle-items-content-main { display: block; position: relative; top: 27px; left: -204px; border-radius: 12px; background: #FFF; box-shadow: 0px 4px 24px 0px rgba(100, 100, 100, 0.25); width: 435px; height: 115px; } .toggle-items-content { display: flex; align-items: baseline; justify-content: center; padding-top: 22px; } .toggle-items-content .items-content-text .h4-saved-story{ color: #000; font-size: 20px; font-style: normal; font-weight: 700; line-height: normal; text-transform: capitalize; margin: 2px 0 10px 6px; } .toggle-items-content .items-content-text .myInc42-plus-dark { width: 100px !important; } .toggle-items-content .items-content-text .myInc42-light { width: 80px !important; } .toggle-items-content .items-content-text img{ height: 22px; } .view-my-feed-btn { width: 100%; text-align: center; display: flex; justify-content: center; } .view-my-feed-btn a { width: auto !important; } .view-my-feed-btn button { border-radius: 4px; background: linear-gradient(180deg, #DA1B4D 0%, #E23026 100%); color: #fff; font-size: 12px; display: inline-block !important; min-width: 162px; width: 162px !important; height: 34px !important; font-style: normal; font-weight: 700; line-height: normal; padding: 10px; cursor: pointer; } .CustomIconStyled { position: absolute; right: 180px; top: -80px; } .SubDropdownModelShare .sub-arrow-icon { width: 76px; height: 80px; position: relative; overflow: hidden; box-shadow: none; } @media (max-width:767px) { .toggle-items-content .items-content-text .h4-saved-story{ margin: 4px 0 10px 6px; font-size: 18px; } .toggle-items-content { align-items: center; } }

When Sarvam AI announced that it was open-sourcing its Sarvam 30B and Sarvam 105B models, it framed the move as a milestone for India’s sovereign AI push. 

Both models were trained from scratch on large datasets curated in-house and built using a mixture-of-experts (MoE) architecture designed to scale reasoning capabilities without dramatically increasing inference costs. Sarvam 30B is positioned as an efficient reasoning model suitable for real-time deployments, while Sarvam 105B is targeted at complex reasoning, coding, and agentic workflows.

The company says the models were trained entirely in........

© Inc42