THE MONTHLY
SIGNAL

Claude Opus 5

Introducing Claude Opus 5

Opus 5 is a step change improvement for the Opus tier powering long-running agents while delivering improvements in coding and professional work.

FLUX 3

FLUX 3: One Multi-Modal Model | Black Forest Labs

One multi-modal model for Image, Video, Audio and Action-Prediction. Creations are truer to life in every kind of style.

GLM 5.2 Vision NVFP4

baseten/GLM-5.2-Vision-NVFP4 · Hugging Face

GLM 5.2 Vision NVFP4 is published on Hugging Face. Open the model or project card for the author’s documentation, files, license and usage details.

GPT Live voice on desktop

ChatGPT Voice | ChatGPT Learn

Try voice in Chat, Work, and Codex in the ChatGPT desktop app.

Health in ChatGPT

Health in ChatGPT links to its original tools page, but the publisher did not expose a readable preview. Open the source directly for the full release.

HOMIE

HOMIE Demo Page

Human-object Centric Video Personalization via Multimodal Cognition Enhancement

Laguna S 2.1

Introducing Laguna S 2.1

Today we’re releasing Laguna S 2.1, a significant step forward in our development of models that pursue longer horizon work and make effective use of reasoning.

Mage-Flow

microsoft/Mage-Flow · Hugging Face

Mage-Flow is published on Hugging Face. Open the model or project card for the author’s documentation, files, license and usage details.

Nanbeige 4.2 3B

Nanbeige4.2-3B

ModelScope——A one-stop service that brings together the most advanced machine learning models from various fields, offering model exploration experience, inference, training, deployment, and application.

OpenAI–Hugging Face security incident

OpenAI–Hugging Face security incident links to its original research page, but the publisher did not expose a readable preview. Open the source directly for the full release.

OpenDreamer

How to train a frontier-level world model

How we trained and open-sourced a frontier-level world model — the lessons, failures, and fixes — with a live, playable demo running on Reactor.

Qwen-Image 3.0

Qwen Studio

Qwen Studio offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.

SANA-Video 2.0

Hybrid Linear Attention with Attention Residuals for Efficient Video Generation.

Self-calibrating quantum computer

Towards a quantum computer that learns from its errors

Volodymyr Sivak and Paul Klimov, Research Scientists, Google Quantum AI, Google Research

ShotPlan

ShotPlan: Cinematic Video Generation with Learnable Planning Token

ShotPlan enables multi-shot cinematic video generation with frame-accurate hard cuts, soft transitions and temporally localized camera movement via learnable planning tokens.

Nunchaku Lite in Diffusers

Bringing Nunchaku 4-bit Diffusion Inference to Diffusers

Diffusers gained native loading for Nunchaku 4-bit diffusion checkpoints, reducing memory while accelerating denoising and removing the need for a separate inference engine.

SymptomAI

SymptomAI: Towards a conversational AI agent for everyday symptom assessment

Google Research evaluated conversational symptom-assessment agents in a national-scale study with 13,917 participants, comparing their differential diagnoses with clinician assessments and wearable biosignals.

Gemini 3.6 Flash family

Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber

We’re introducing new Gemini models, including Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber.

Grabette

Grabette: an open system to record robot-manipulation data

Pollen Robotics open-sourced a handheld system for recording manipulation demonstrations and turning them into LeRobot datasets through a browser-based processing pipeline.

Cosmos 3 Edge

Introducing Cosmos 3 Edge

NVIDIA released a 4B open world model for on-device physical AI, combining visual understanding, prediction and robot-action generation for edge hardware including Jetson.

Ardy

ARDY: Autoregressive Diffusion for Interactive Motion

ARDY is an autoregressive diffusion model for interactive human motion generation with online text prompting and flexible kinematic constraints.

GenCeption

Video Generation Models are General-Purpose Vision Learners

GenCeption: one unified, feed-forward vision model built on video generative pretraining — depth, normals, pose, segmentation, keypoints & 4D grounding with SOTA performance, steered by text instructions.

GNM

GitHub - google/GNM: An open ecosystem of parametric human models and perception stacks, starting with GNM Head.

An open ecosystem of parametric human models and perception stacks, starting with GNM Head. - google/GNM

Kimi K3

Kimi K3 Tech Blog: Open Frontier Intelligence

Kimi K3 is the world's first open 3T-class model — frontier performance across coding, knowledge work, and reasoning, with native multimodality and 1M context.