Google DeepMind's DiffusionGemma generates 256 tokens in parallel, hitting 1,100+ tokens/sec on a single H100 — a new architecture for speed-first local inference
Ion Stoica, the Berkeley professor behind Databricks, Anyscale, and Chatbot Arena, joins Liquid AI's advisory council to help scale its hardware-aware, on-device AI platform.
Google's Magika uses a 3MB deep learning model to detect 200+ file types in 5ms, already guarding Gmail and Google Drive at hundreds of billions of files per week
Vibe-Trading is an open-source multi-agent framework that turns plain-English prompts into backtested trading strategies, now with live broker execution across 10 platforms
Biohub open-sources ESMC, ESMFold2, and a 6.8B-protein Atlas — beating AlphaFold 3 on antibody-antigen prediction and designing lab-validated cancer binders in days
NVIDIA's NVFP4 quantization squeezes Google's text-diffusion model into 18GB VRAM, hitting 1,100+ tokens/sec on a single H100 — a new speed ceiling for local inference
Mechanize's GBA Eval benchmark shows Claude Fable 5 hitting 74.5% — beating Opus 4.8's 24-hour score in under 2 hours, with a notable audio regression.
Prime Intellect applies ECHO to teach LLMs to predict their own tool outputs during RL, improving agent generalization and efficiency at near-zero extra cost.
GitHub drops the waitlist for its Copilot desktop app, opening agent-native development to all paid Copilot subscribers with new canvases, Agent Merge, and cloud sessions.
Google DeepMind's randomized controlled trial in Sierra Leone shows Gemini's Guided Learning mode delivered up to 2.5 years of math progress in just 8 weeks