
Apple's MLX Turns Four Macs Into a Trillion-Parameter Cluster
MLX's JACCL backend brings RDMA over Thunderbolt 5 to Mac clusters, delivering sub-50µs latency and up to 3x LLM inference speed across four Macs
20 stories

MLX's JACCL backend brings RDMA over Thunderbolt 5 to Mac clusters, delivering sub-50µs latency and up to 3x LLM inference speed across four Macs

Hugging Face's speech-to-speech library gives you a fully local, OpenAI Realtime-compatible voice agent pipeline where every component is swappable and free

Kilo ran a controlled code-review benchmark on GLM-5.2 and found prompt wording matters more than reasoning effort — with a clear local-vs-cross-file split

HolyClaude packages Claude Code, 8 AI CLIs, a headless browser, and 50+ dev tools into a single Docker container that spins up in one command

Sakana AI's Sheaf-ADMM fuses distributed optimization theory and algebraic topology to build transparent, coordination-native neural networks accepted at ICML 2026

Superpowers 6 adds Kimi Code support and slashes build costs by 60% using Anthropic's Fable to optimize its own workflow overnight

vLLM-Omni's staged pipeline engineering cuts Qwen3-Omni's first audio latency from 6s to 0.6s and boosts throughput 5.4x on the same GPUs

Tongyi Lab's weekly digest bundles a head-level hybrid attention architecture, community NVFP4 quants for Qwen3.6-27B, and the story behind LOGOS.

Sakana AI unifies two competing families of gradient-free optimizers into one framework, then uses the insight to merge LLMs cheaper and better.

Alibaba's Wan Video adds music-driven dance generation: upload a character image, pick a style, and get a beat-synced video in seconds

Ouroboros is an open-source Agent OS that interviews you before writing a single line of code, using math to block premature builds until your spec is clear enough.

Sakana AI's Ren Ito joins 40+ global tech and government leaders on the newly formed UN-ITU AI for Good Global Commission

Anthropic raises Claude API rate limits across all tiers and drops spend-based tier progression, giving Sonnet and Haiku users up to 5x more throughput

NVIDIA cut the first tagged release of its open-source TensorRT Cookbook, updating all examples to TensorRT 10.16 and adding a new TensorRT-RTX section.

Liquid AI's LFM2-VL-450M brings multimodal vision to edge devices with 2x GPU inference speed and under 500M parameters

Liquid AI's LFM2.5-350M packs 28T-token training and reinforcement learning into a sub-1GB model built for edge agentic pipelines

AI-powered vulnerability hunting by Anthropic and OpenAI triggered a 3.5x spike in high-severity CVE disclosures, reshaping the entire cybersecurity landscape

Runway adds slash-command Skills to its AI Agent, letting marketers build full ad campaigns, commercials, and localized assets in a single command

Pika's new Voxel-It Skill turns any photo into a Minecraft-style world, while its MCP server brings the full creative suite into Claude, Codex, and more

Fireworks AI ships a new serving tier for Z.ai's GLM 5.2 that runs 2-3x faster than its Standard path with no quality drop.