TurboQuant+ brings Google's ICLR 2026 KV cache compression to Apple Silicon and vLLM, squeezing 4.9x more context into the same memory with near-zero quality loss
Google DeepMind's first-ever robotics accelerator gives 15 European startups access to Gemini models, AI infrastructure, and up to $350K in cloud credits — equity-free.
MemPalace hits 96.6% on LongMemEval with zero API calls, using verbatim storage and local vector search — no cloud, no subscription, no LLM at write time
Artificial Analysis swaps SWE-Bench Pro for DeepSWE in its coding agent index, reshuffling the leaderboard and exposing how gameable the old benchmark had become
PaddleX 3.x is Baidu's open-source all-in-one AI toolkit with 270+ models, 33 ready-to-run pipelines, and SOTA document parsing — all behind a single Python API
Reasonix is a DeepSeek-native terminal coding agent that slashes long-session API costs to ~1/5 by engineering every layer around prefix-cache stability
GitHub's new standalone desktop app drops the waitlist, letting paid Copilot users run parallel AI agent sessions from issue to merged PR in one place.
Replit and Databricks expand their integration with per-user data permissions in public preview, letting anyone build apps where each viewer only sees data they're authorized to access
Google Colab adds Gemini-powered Visualization Mode and one-link Fullscreen Sharing, turning notebooks into shareable dashboards without writing chart code
Cohere's first open-source coding model, North Mini Code, is a 30B MoE with only 3B active parameters — and the community already ported it everywhere in 48 hours.