Training a 2.7B MoE from scratch for $200, one GPU at a time
A community project trained NanoColibri-Instruct, a 2.7-billion-parameter mixture-of-experts model, for a total cost of roughly $180-260 using a “relay pretraining”…
A community project trained NanoColibri-Instruct, a 2.7-billion-parameter mixture-of-experts model, for a total cost of roughly $180-260 using a “relay pretraining”…
Bekko Embedding is a family of compact multilingual retrieval models built through layer pruning and contrastive learning. Its smaller variant,…
Liquid AI released two encoder models, LFM2.5-Encoder-230M and -350M, built for efficient long-context processing on CPU by converting decoder architectures…
OpenAI described engineering work, partly performed autonomously by its GPT-5.6 “Sol” model, to optimize the production software that serves its…
The VIDRAFT team documented its winning entry to the Fast Gemma Challenge, reaching 510.58 tokens per second at a perplexity…
The FCC announced on July 28, 2026 that it will bar imports of new Chinese-made humanoid and quadruped robots, along…
Jacob Tsimerman, a University of Toronto mathematician who won the 2026 Fields Medal for his work on the Andre-Oort conjecture,…
OpenAI announced on August 1, 2026 that an internal version of its upcoming Astra model family, designed to coordinate multiple…
Palantir reported second-quarter 2026 earnings of 41 cents per share on revenue of $1.94 billion, both ahead of analyst expectations…
Meta and Newsmax announced an AI content-licensing partnership on July 28, 2026, giving Meta rights to train AI search and…
The US Commerce Department signed a letter of intent to give GlobalFoundries $300 million in CHIPS Act funding in exchange…
Autodesk completed its all-cash acquisition of maintenance-management software provider MaintainX on August 3, 2026, its largest deal ever. The company…