News
Curated articles, my blog posts, and site announcements.
Curated articles, my blog posts, and site announcements.
EXO announces a benchmarking effort covering every model, quantization, and hardware setup at every price point, to be released publicly so developers can pick working configurations.
EXO refreshes its site, opens a public Discord, and kicks off weekly community education sessions starting May 24.
EXO v1.0.71 lands with better sampling defaults, M5 Mac and RDMA bug fixes, and day-zero support for Moonshot's Kimi K2.6.
EXO v1.0.70 ships multimodality, major long-context memory improvements for coding agents, and expanded model support.
EXO ships day-zero support for MiniMax M2.7 with RDMA and tensor parallelism for near-linear scaling across Mac clusters.
EXO v1.0.69 turns continuous batching on by default and adds support for M5 Pro and M5 Max chips.
EXO Labs publishes its founding thesis tweet positioning open-source decentralized infrastructure as the future of AI.