Edge0-145B: Open-Source On-Device AI with 145B Parameters

Why it matters

Edge0 has released two open-weight models, Edge0-145B and Edge0-120B, along with the inference framework and deployment tools needed to run them, positioning large-scale AI directly on personal hardware rather than in the cloud. The company frames the release as an answer to a longstanding tradeoff: capable models have demanded cloud infrastructure, while local models have historically been small and limited. The technical explanation centers on co-designing the model and its engine. Large matrices are shared across groups of sparse experts, so only a small fraction of total parameters is active at each step. Training was conducted in the on-device architecture from the start rather than compressing a cloud model afterward. Loading pipelines work across GPU memory, system memory, and the SSD, while scheduling splits computation between CPU and GPU and uses speculative decoding. The result, per the company's internal testing, is a model with frontier-cloud-level coding, reasoning, and agent capabilities running on a laptop. The release matters because it challenges the assumption that serious AI work requires cloud dependence, fees, and connectivity.