Quanta AI Labs

49 / Dec 1 – Dec 7, 2025 · 13 RELEASES

AI News Week 49, 2025

Claude Opus 4.5 and Z-Image headline frontier models, while GeoVista, Fara-7B, and RynnVLA push agentic AI forward.

Language Models (3)

  • Claude Opus 4.5

    Anthropic · Reasoning

    Anthropic's most capable model yet, scoring higher than any human on performance engineering exams with improved vision, reasoning, and mathematics.

  • DeepSeek-Math-V2

    DeepSeek AI · Reasoning

    Self-verifiable mathematical reasoning model with improved evaluation results across mathematical benchmarks.

  • ChatGPT Shopping

    OpenAI · Text Generation

    ChatGPT shopping research feature enabling product discovery and comparison directly in conversations.

Vision & Image (5)

  • FLUX.2 (Official)

    Black Forest Labs · Text-to-Image

    Official FLUX.2 release from BFL with frontier visual intelligence, improved prompt following, and open-core availability.

  • Z-Image

    Alibaba Tongyi · Text-to-Image

    Efficient 6B-param image generation foundation model with photorealistic quality and bilingual text rendering comparable to commercial models.

  • Z-Image (GitHub)

    Tongyi-MAI · Text-to-Image

    Open-source single-stream diffusion transformer with Decoupled-DMD acceleration and DMDR reinforcement learning fusion.

  • Hunyuan Vision

    Tencent · Visual Understanding

    Tencent's multimodal vision platform with cross-domain knowledge and natural language understanding capabilities.

  • iMontage

    NTU / Tencent · Text-to-Image

    Unified many-to-many image generation with versatile in-context learning, style transfer, and storyboard creation.

Agents & Automation (4)

  • GeoVista

    Tsinghua / PKU · Web Agents

    Web-augmented agentic visual reasoning for geolocalization with image zoom-in and web-search tool integration.

  • Fara-7B

    Microsoft Research · Desktop Agents

    Efficient 7B agentic model for computer use with robust safety measures, matching larger systems on web tasks.

  • RynnVLA-002

    Alibaba DAMO · Embodied AI

    Unified Vision-Language-Action and World Model combining VLA for action prediction and world model for video prediction.

  • AlohaMini

    liyiteng · Robotics

    Open-source dual-arm mobile robot with motorized lift for robotics research and embodied AI development.

Developer Tools (1)

  • ADE Python SDK

    Landing AI · Data Integration

    Python SDK for document parsing and extraction with async support, file uploads, and error handling.

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