
Microsoft's MAI-Image-2.5 Hits No. 2 on Leaderboard With Precise Localized Editing
Microsoft's MAI-Image-2.5 ranks No. 2 for image editing on Arena, with a unified generation and editing API now live on Foundry and OpenRouter
20 stories

Microsoft's MAI-Image-2.5 ranks No. 2 for image editing on Arena, with a unified generation and editing API now live on Foundry and OpenRouter

Meta's SAM 3D earned a CVPR Best Paper Honorable Mention for turning single 2D photos into full 3D objects, scenes, and layouts at unprecedented quality.

Meetily is a fully local, open-source meeting assistant that transcribes and summarizes meetings on your own hardware using NVIDIA's Parakeet or Whisper, with zero data leaving your device
AlloGen introduces a learned scoring network that steers any protein binder generator toward molecules selective for one conformational state.

Ideogram releases its first open-weight model: a 9.3B single-stream DiT that beats much larger rivals on typography and design tasks.

Unsloth's quantization-aware training GGUFs squeeze Google's 26B Mixture-of-Experts model into 15GB while keeping near-bfloat16 accuracy intact.
MIT researchers built an AI scientist that uses category theory to mathematically separate genuine discovery from mere search, with measurable novelty.

Microsoft ships seven in-house AI models at Build 2026, from a frontier reasoning model to voice and transcription, all trained without OpenAI data

Prime Intellect joins NVIDIA's Nemotron Coalition, contributing 2,500+ open RL environments to turn Nemotron 3 Ultra into a fully customizable open agent platform

Alibaba's Tongyi Lab ships a transparency-focused overhaul of its agent framework with observable events, retry logic, granular permissions, and pluggable workspaces.

Liquid AI's new Extract models take an image and a YAML field list, return clean JSON, and run on-device with no parsing layer.

Researchers from the University of Tokyo show that swapping hard token assignment for soft distributions only at inference time recovers lost speech information without retraining

VoxCPM2 is a fully open-source 2B-parameter TTS model covering 30 languages, voice cloning, and natural-language voice design — no discrete tokenizer required

A new scaling law and masking trick show how to squeeze more out of repeated training data as the internet runs dry

A new theoretical framework proves why training on more diverse tasks makes transformers learn faster and generalize further — explained through low-dimensional geometry.

A new study tests 40+ speech models against human voice similarity judgments, revealing which architectures and layers actually match how humans hear voices.

Microsoft puts Dev Box into maintenance mode and pushes Windows 365 as the standard cloud workstation, shipping a preconfigured developer image in public preview.

Codex now has a GitHub-style profile page tracking streaks, lifetime tokens, and top features, with optional shareable cards for your activity.

Google launches Managed Agents in the Gemini API, letting developers spin up sandboxed, tool-using agents on Google's infrastructure with one call.

NotebookLM's new Source Attribution feature exposes the exact prompt and sources behind every Studio artifact, and lets you tweak and re-run them.