48 / Nov 16 – Nov 22, 2026 · 3 RELEASES

AI News Week 48, 2026

Lattice And Game Agents. OpenAI published "A step toward defining the Standard Model on a lattice", which describes a candidate nonperturbative lattice definition of the Standard Model and says GPT-6 Astra helped turn the idea into a detailed proof. The page gives no measured results or availability details. Run Luo and seven co-authors released "Gamer: A Generalist Vision-Language-Action Foundation Model for Long-Horizon Real-Time Game Agent" on arXiv. It is pre-trained on tens of thousands of hours of keyboard and mouse trajectories, then tuned with sparse-thinking instruction tuning and multi-turn reinforcement learning, and the authors report solid generalisation across several 3D open-world games. Bland released "Tofu-1", a full-duplex speech-to-speech model aimed at noisy, high-stakes phone calls, in an enterprise research preview. No benchmarks, model size or pricing have been published for it.

Audio & Speech (1)

  • Tofu-1

    Bland · Full-Duplex Speech Model

    Bland has released Tofu-1, a full-duplex speech-to-speech model aimed at noisy, high-stakes phone calls. It is available in an enterprise research preview. No benchmarks, model size or pricing have been published yet.

    Open the source ↗

Agents & Automation (1)

  • Gamer: A Generalist Vision-Language-Action Foundation Model for Long-Horizon Real-Time Game Agent by Run Luo, Kuan Zhang, Qi Deng, Yukun Chen, Zhihao Yang, Yufei Zhang, Renrong Hu, Yi Zhang (arXiv)

    Gamer: A Generalist Vision-Language-Action Foundation Model for Long-Horizon Real-Time Game Agent

    Run Luo, Kuan Zhang, Qi Deng, Yukun Chen, Zhihao Yang, Yufei Zhang, Renrong Hu, Yi Zhang (arXiv) · Game VLA Agent

    Gamer is a generalist agent framework for long-horizon tasks in interactive environments, combining perception, memory management and multi-step reasoning. It is pre-trained on tens of thousands of hours of device-level keyboard and mouse trajectories from simulated games and embodied environments, then tuned with sparse-thinking instruction tuning and multi-turn reinforcement learning. The authors report solid generalisation across several 3D open-world games, efficiency comparable to skilled players, and competitive results on conventional 2D game benchmarks.

    Open the source ↗

Research & Safety (1)

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.