June 26, 2026
Liquid AI LFM2.5 230M runs at 1,400 tok/s entirely in the browser via WebGPU—faster than any cloud API, completely private, and costing exactly $0. The future of AI inference is local.
June 26, 2026
A head-to-head comparison on reproducing the SDPO paper with full ablations. GLM 5.2 completes the task for $6.98 (6.6x cheaper than Opus 4.8) while using 41% fewer tokens. The era of affordable AI-assisted research is here.
June 26, 2026
DeepReinforce releases Ornith-1.0—a family of open-source coding models (9B to 397B MoE) that learn to generate their own task-specific scaffolds during RL training. State-of-the-art results on Terminal-Bench, SWE-Bench, and more.
April 19, 2026