Anthropic launches Claude Fable 5 visual AI model amid broader AI model breakthroughs

Anthropic announced Claude Fable 5, the first publicly released Mythos‑level AI model with visual‑state‑of‑the‑art capabilities. The model can reconstruct an entire web‑application’s source code from a single screenshot and claims top performance on visual benchmarks. The U.S. government has restricted foreign access to the model citing national‑security concerns.

RoboScience unveiled its Visics general‑embodied model, built on a Vision‑Language‑Object‑Action (VLOA) architecture that uses a unified Object Trajectory representation. Visics demonstrates cross‑robot, cross‑object and cross‑task abilities such as furniture assembly and dexterous grasping, aiming to overcome the generalisation limits of previous embodied AI systems.

Chinese startup Ying Mou Technology completed a multi‑billion‑RMB financing round and released Hyper3D Rodin Gen‑2.5, the first 3D generative model with million‑face output and a “think‑then‑generate” workflow. The model delivers production‑ready 3D assets at adjustable compute budgets, and has been integrated into platforms from Nvidia to Unity and Canva.

SenseTime is developing a multimodal design‑focused model codenamed U1 Pro, intended to rival OpenAI’s GPT‑Image 2. The model supports 8K resolution outputs and will begin internal testing in July, aiming to provide “thinking” capabilities for complex design tasks.

NOETIX Robotics showcased the Hobbs 3 bionic robot head, featuring 32 active and 8 passive degrees of freedom, platinum‑silicone skin for wax‑like realism, integrated RGB cameras, microphones and speakers, and low‑latency interaction. The system targets commercial applications such as reception, live‑stream interaction and home companionship.