New releases

The General Data Corporation · General Data's open PII detection model built for Europe: it tags 20 kinds of personal data, including national ID formats (French NIR, German Steuer-ID…

Microsoft · Microsoft Research AI for Science's foundation model for biomolecular structure, dynamics and function, in public preview

Anam · Anam's real-time avatar model, built on Cara-4 to be faster and more expressive, with better lip sync, Director Notes style prompts and wider framing

NVIDIA · NVIDIA's 16B surgical-robotics model for dVRK-style and other surgical robots

Deployxa · Deployxa's 1.5B open-weights SRE model that diagnoses container and deployment failures and returns machine-executable JSON fix recipes

ElevenLabs · ElevenLabs' new class of speech-to-speech models for voice agents, which the lab says starts talking in about half a second

Hao AI Lab (UC San Diego) · Hao AI Lab's FastVideo team's 8-step reference-to-video model, distilled from MiniMax's MiniMax-H3: video with synchronized audio from a text prompt plus…

Bria AI · Bria AI's image generation and editing release, trained only on licensed data (lab statement)

NVIDIA · NVIDIA GEAR lab's suite of open models that watch robot and human video and turn it into training signals: timestamped captions, dense progress rewards and…

Maximo AI · Maximo AI's first sparse model, aimed at agentic coding, UI design and real-time 3D from a single prompt

MemTensor · MemTensor's open-weight model family (Flash and Nano) with persistent memory built into model computation, so agents can reuse past history without reloading…

Africa Compute Fund · Africa Compute Fund's general-purpose LLM with an emphasis on economic, institutional and development knowledge, including African economies, languages and…

Oogam AI · Oogam AI's speech-to-text and text-to-speech models for 28 Indian languages (3 in beta), offered through the Oogam API

Zenithon · Zenithon calls NOVA the first world model for extreme physics: a single transformer pretrained on plasma and fluid physics systems that predicts how complex…

Radical Numerics · Radical Numerics' genome language model, pretrained across 28 biological modalities including DNA, RNA, proteins and epigenomics, and adapted to eight…

Ornith · Ornith's 122B mixture-of-experts model for coding and agentic work, with open weights under the MIT license

Gradium · Gradium's suite of on-device text-to-speech models under 50M parameters, built to run offline on phones, laptops, small boards and in the browser

Kyutai · Kyutai's Pocket TTS v4.0.0 release: six small English text-to-speech models with voice cloning, at 18.2M, 6.9M and 1.9M parameters, each with a…

NVIDIA · NVIDIA and the Baker Lab's flow-matching model for biomolecular design

Microsoft · Microsoft's 7B open-weights model, fine-tuned from Mistral-7B-Instruct-v0.2, that labels a user message about a service, product or agent as Negative, Neutral…

VideoGen · VideoGen's video model that turns a brief, with optional reference images, into a finished multi-shot video with audio, up to ten minutes long (lab statement)

Mistral · Mistral's new speech-to-text and text-to-speech models, with a voice agent SDK built on LiveKit and Pro Voice cloning from up to 30 minutes of audio

SenseTime · SenseTime's 27B embodied model for robot planning and decision-making, trained from Qwen3.8-27B with experience from its open-source SenseNova-RoboRSI harness…

Bindwell · Bindwell's foundation model for predicting experimental outcomes in molecular discovery, trained across 800M–9B sizes on about 80M experimental measurements