Indian artificial intelligence company Sarvam AI announced at the inaugural Sarvam Epoch conference, which opened this Thursday local time, that it will develop an artificial intelligence model with parameters numbering in the trillions.
Sarvam AI said that its existing 105B model performs excellently in many areas, while its speech model offers significantly better cost-effectiveness than global competitors.
The company plans to develop a trillion-parameter model because of its belief that “India should produce the tokens it consumes,” and that India should avoid “double-paying” foreign AI companies with both money and data. The envisioned new model is expected to be completed within the next six months and could compete with leading models in areas such as programming, cybersecurity, and scientific research.
Sarvam AI announced yesterday that it will establish an office in the United States to attract Indian-origin AI talent there to return. It also announced two model products: the Bulbul V4 text-to-speech model and the Saaras V4 Multi-speaker speech recognition model.