Sarvam AI has launched Sarvam Vision, a 3B-parameter state-space vision-language model that outperfo...
The AMW Read
The launch of a specialized, high-performing vision-language model for Indic languages updates the player map and signals a shift toward sovereign AI capabilities to counter global model dominance in specific geopolitical regions.
NoveltySignificance
Foundation Models Β· Player MapGeopolitics
Sarvam AI has launched Sarvam Vision, a 3B-parameter state-space vision-language model that outperforms Google Gemini and ChatGPT in India-specific benchmarks. The model achieved 93.28% accuracy on OmniDocBench v1.5, demonstrating superior handling of 22 official Indian languages and complex document layouts. This marks a critical milestone for sovereign AI, enabling high-precision digitisation for local government and financial sectors that global models struggle to support. By prioritizing Indic-first architecture, Sarvam is effectively closing the regional accuracy gap in the global AI race. π
