🤖 AI Weekly Recap (Week 5)

Plus: The most important news and breakthroughs in AI this week

Happy Sunday! We just had another crazy week in AI. Moonshot AI just dropped the world’s most advanced open-source AI model while a new AI lets you create unlimited motion graphics in seconds with just a click.

And that's not all, here are the most important AI moves you need to know this week.

Prism is a free, web-based scientific workspace available to anyone with a ChatGPT account. Built on GPT-5.2, it works like an AI-native word processor for research papers, helping scientists write, review, and reason more effectively, without replacing human judgment.

  • AI-assisted writing, claim checking, and literature search inside a single workspace

  • Full project context awareness, so the model understands your entire paper

  • Native LaTeX support plus AI-generated diagrams from sketches

  • Designed to accelerate human scientists, not automate research end-to-end

InVideo unveiled a new AI Motion Graphics feature built with Anthropic, designed to let anyone create studio-quality animations using simple text prompts. Think “vibecoding,” but for motion design.

  • Prompt-based creation: describe the animation in natural language, AI handles the rest

  • Zero learning curve: no keyframes, layers, templates, or coding needed

  • Multi-agent AI system plans, scripts, and generates visuals automatically

  • Exports high-quality motion graphics up to 4K resolution

Try it Now → https://invideo.io/


Genie 3 is a DeepMind research prototype that turns text prompts (and images) into explorable 3D worlds, essentially letting you create a video game on the fly. It’s powered by Gemini, Nano Banana Pro, and Veo 3, and is still labeled experimental.

  • Generate interactive 3D worlds from text prompts and images

  • Explore worlds in real time with adjustable camera angles and perspectives

  • Remix existing worlds or build on shared prompts from a curated gallery

  • Download videos of your world explorations (with current 60-second limits)

Alibaba Cloud’s Qwen team unveiled Qwen3-Max-Thinking, a proprietary reasoning model built around test-time scaling and autonomous tool use. Instead of brute-force sampling, it iteratively reflects, redirects compute, and grounds answers with live tools.

  • Scores 49.8 on Humanity’s Last Exam (with search), beating GPT-5.2-Thinking and Gemini 3 Pro

  • Uses test-time scaling to trade compute for intelligence without exploding token costs

  • Natively combines reasoning + tools (web search, memory, code interpreter) in one turn

  • Aggressively priced at $7.20 / 1M tokens, undercutting most flagship reasoning models

Try it Now → https://chat.qwen.ai/

Lucy 2 is a persistent world model that generates video frame by frame while a camera stays live, enabling always-on generative video with consistent motion, identity, and physics.

  • Runs continuously at 1080p, 30 FPS with no buffering, resets, or time limits

  • Maintains consistent bodies, lighting, clothing, and environments in real time

  • Cuts real-time generation costs to ~$3/hour, down from hundreds

  • Already tested with Twitch creators and integrated into NVIDIA Omniverse for robotics sims

Kimi K2.5 is Moonshot AI’s latest open LLM, built on a mixture-of-experts design and trained on 15 trillion tokens, including multimodal data. The company says it’s optimized for faster reasoning, agent-based workflows, and lower hardware usage.

  • 1T-parameter MoE model with only ~32B active parameters per task

  • Trained on 15T tokens (text + multimodal), improving chart and file understanding

  • “Thinking mode” + Agent Swarm can orchestrate up to 100 agents per prompt

  • Outperformed GPT-5.2 on HLE-Full and matched top models across 20+ benchmarks

Try it Now → https://www.kimi.com/

Thanks for making it to the end! I put my heart into every email I send. I hope you are enjoying it. Let me know your thoughts so I can make the next one even better.

See you tomorrow :)

Dr. Alvaro Cintas