India's Sphota AI Model Claims 50x Training Efficiency, 4.5x Memory Gains Over Liquid AI

Indian AI startup NeuroDX has unveiled Sphota, a 240-million-parameter model that it says outperforms a comparable Liquid AI model on efficiency metrics, signaling a major step for India's frontier AI ambitions. The company claims Sphota achieves 50 times greater training data efficiency and 4.5 times better memory efficiency, potentially lowering the cost and compute barriers for advanced AI development. These gains could help Indian firms compete globally without relying on massive datasets or expensive infrastructure. The announcement comes as India pushes to build sovereign AI capabilities amid rising demand for efficient, scalable models. NeuroDX has not yet released full technical details or independent benchmarks, but the claims, if verified, could position Sphota as a template for resource-constrained AI innovation. Industry observers note that efficiency breakthroughs are critical for deploying AI in edge devices and emerging markets. The model's performance against Liquid AI, a U.S.-based startup known for liquid neural networks, adds competitive pressure in the race for next-generation architectures. For India, Sphota represents both a technical milestone and a strategic asset in the global AI landscape. The next steps will be independent validation and real-world deployment to prove the model's practical value beyond lab metrics.