SiMa.ai introduced Palette 3.0 for deploying multiple AI models on Modalix chips. It supports container-based configurations and validation using sensor simulation, focusing on combining language, vision and other models in edge systems. The company cites up to a 3.5-fold speedup from speculative decoding and an example running GPT-OSS 20B on a 16 GB module. These figures are not guarantees for every model or input length, so actual memory use, latency and performance under shared resources require separate evaluation.