· Updated
Bonsai 8B, the first commercially viable 1-bit Large Language Model. While standard models use 16-bit or 8-bit weights, this architecture compresses weights to a single bit. The result is a model 14x smaller and 8x faster than traditional counterparts.This release addresses hardware bottlenecks that prevent frontier-level intelligence from running locally on smartphones. By maintaining accuracy on a 1-bit Pareto curve, Bonsai 8B enables complex reasoning on low-power chips with 5x better energy efficiency. It removes the need for expensive cloud GPUs or constant connectivity for high-performance tasks.
You can now deploy an 8B-parameter model on consumer hardware that previously lacked the VRAM for local AI. The Bonsai series is open-sourced for integration into privacy-sensitive applications. This makes high-speed, on-device AI practical for mobile and embedded systems without sacrificing the model's underlying intelligence.
Today, we are emerging from stealth and launching PrismML, an AI lab with Caltech origins that is centered on building the most concentrated form of intelligence. At PrismML, we believe that the next major leaps in AI will be driven by order-of-magnitude improvements in https://t.co/QzAJDIyFWK
More like this
Cohere10h ago
OpenRouterApr 21
CloudflareApr 19