SiMa.ai has released Palette 3.0, updated software for its Modalix edge AI chips, aiming to let multiple AI models run at the same time on a single accelerator rather than assigning hardware to one workload. The company positions Palette 3.0 as making the chip behave more like a “shared workstation,” using adjustable priorities across containers and hardware engines.

Across coverage, SiMa.ai’s main performance claim is faster text generation, with reports citing up to 3.5x higher throughput on language models through speculative decoding, while power draw is said to stay broadly the same or energy per request to be reduced. The company also describes support for concurrent language and vision workloads for use cases such as robots, drones, industrial cameras and autonomous vehicles.

Outlets differ in emphasis and included details: some focus on the throughput uplift and speculative decoding mechanism, while others add specific experimental framing (such as many simultaneous video streams) and product-adoption notes (recompilation) or pricing information for developer kits.