
NVIDIA Upgrades nccl-tests to Benchmark GPU-Initiated Networking on Multi-GPU Clusters
NVIDIA updates nccl-tests with flags to benchmark the new device-side communication APIs and symmetric memory introduced in NCCL 2.28+
20 stories

NVIDIA updates nccl-tests with flags to benchmark the new device-side communication APIs and symmetric memory introduced in NCCL 2.28+

Google's Gemini models are now callable inside Apple's Foundation Models framework via Firebase, letting iOS developers swap between on-device and cloud AI with one line of Swift code.
NVIDIA's NVFP4 training recipe for JAX and MaxText delivers up to 1.73x faster pretraining than FP8 on Blackwell GPUs, with no measurable accuracy loss.
OpenEnv gains a multi-org governance committee spanning Meta, NVIDIA, Hugging Face, and more — becoming the shared infrastructure layer for open-source agentic RL training.
GitHub's experimental Copilot accessibility agent has reviewed 3,535 PRs with a 68% resolution rate — here's the multi-agent architecture behind it
Kaggle now lets users mint permanent DOIs for competition writeups, making ML solutions citable in academic papers for the first time.

NotebookLM gets Gemini 3.5, a sandboxed cloud computer, and 11 export formats — turning a document Q&A tool into an agentic research workstation
xAI's grok-imagine-video-1.5-preview lands #2 on the image-to-video leaderboard with native synchronized audio baked into every generation
Artificial Analysis benchmarks MiniMax-M3 at 55 on its Intelligence Index — set to be the top open-weights model once weights ship, which hasn't happened yet
Cognition's FrontierCode benchmark asks whether AI code is actually mergeable, and today's best models score just 13.4% on the hardest tasks

SakanaAI's open-source ShinkaEvolve uses LLM-guided evolution to discover state-of-the-art algorithms with orders of magnitude fewer evaluations than AlphaEvolve.
nesquena's community-built web UI for Nous Research's Hermes Agent surges 30% in a week after a contributor-driven visual overhaul and new onboarding flow
Vercel's v0 Max tier upgrades to Claude Opus 4.8, bringing Anthropic's most capable public model to AI-powered UI generation

Zed's file finder now lets you open multiple files in a row without closing the prompt, using a single right arrow key press.
Perplexity and Harvard publish the first large-scale study showing AI agents cut task time by 87% and cost by 94% versus search-assisted humans
NotebookLM gets Gemini 3.5, a cloud code sandbox, agentic web search, and a full suite of exportable output formats — all rolling out to Google AI Ultra subscribers now
vLLM-Omni v0.22.0 adds day-0 NVIDIA Cosmos 3 support, robot serving APIs, production TTS, and broader quantization across NVIDIA Blackwell, AMD ROCm, Intel XPU, and Ascend NPU.

Runway's Aleph 2.0 can now reformat any video for any aspect ratio, intelligently filling in the scene rather than just cropping it

Manus now lets you connect multiple Gmail and Google Calendar accounts, letting its AI agent act from the right identity across work, personal, and client workflows
Moonshot AI's Kimi Work brings 300 parallel agents, browser automation, and built-in finance data to your local desktop — no API wrangling required.