Sarvam AI has launched Sarvam Vision, a 3B-parameter state-space model for multilingual document int...
The AMW Read
Updates the player map with a specialized regional model and validates the Frame 2 (CN/OSS/Sovereign) pattern of using efficient architectures to challenge frontier generalists in localized contexts.
NoveltySignificance
Foundation Models Β· Player MapGeopolitics
Sarvam AI has launched Sarvam Vision, a 3B-parameter state-space model for multilingual document intelligence across 22 Indian languages. It outperforms frontier models like GPT-5.2 and Gemini 3 Pro in local OCR and semantic extraction tasks. By unlocking knowledge from vast physical archives, it establishes a foundation for sovereign AI in research and governance. This move highlights the shift toward hyper-localized, efficient architectures that beat generalist models in regional contexts. π
