Revolutionizing AI: PrismML’s Groundbreaking LLM Technology

The Rise of PrismML: Revolutionizing AI with Groundbreaking Compression Technology

PrismML, the AI lab that is quickly making a name for itself in the tech industry, is not just another startup chasing after funding. Led by a team of technical minds with a vision, PrismML is developing industry-changing technology that is set to redefine the capabilities of reasoning large language models.

One of PrismML’s key beliefs is that high-performing language models don’t necessarily have to be large in size. In fact, the startup is on a mission to make reasoning models so compact that they can easily fit on PCs and smartphones. This innovative approach has caught the attention of industry giants, with rumors swirling about potential collaborations with Apple.

Recently, PrismML unveiled its latest achievement – Bonsai 2 27B, a model that compresses Qwen3.8 27B, a widely used open-source model, down to a mere 5.9 GB. This significant reduction in memory size, from the original model, opens up a world of possibilities for deploying advanced AI models on a wider range of devices.

Founded by a group of Caltech researchers and spearheaded by CEO Babak Hassibi, a renowned expert in compression technologies, PrismML is backed by notable investors such as Khosla Ventures, Cerberus Capital, and Caltech. The startup’s advisory board includes Ion Stoica, a co-founder of Databricks and director of Berkeley’s Sky Computing Lab.

While PrismML is not the only player in the field of LLM compression tech, its unique approach sets it apart from the competition. The startup’s compression technology has proven to retain the performance of the original models, with Bonsai 2 matching 98% of Qwen’s aggregate benchmark scores.

PrismML’s success lies in its ability to shrink the “weights” of a model, simplifying the information learned during training into ternary values of +1, -1, or 0. This groundbreaking technique drastically reduces the space required to store model data, paving the way for more efficient and compact AI models.

Looking ahead, PrismML aims to apply its compression technique to even larger models in the future. With plans to release models in the several-hundred-billion-parameter range, the startup is confident that it can maintain intelligence while reducing size.

By enabling advanced models to run on users’ devices, PrismML is ushering in a new era of accessible and private AI technology. As Ion Stoica puts it, “You are going to have intelligence at your fingertips, and it’s going to be free because it’s going to run on the device you already bought. It’s also going to be private, because you’re not going to send it to the cloud.”