🤖 AI Weekly Recap (Week 6)

Happy Sunday! We just had another crazy week in AI. OpenAI just dropped a desktop app that runs multiple AI agents working in parallel on your projects while new AI lets you create publication-ready research papers with professional charts in one prompt.

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

1. Kling 3.0 Levels Up AI Video Creation

Kling 3.0 is positioned as an “all-in-one creative engine” for multimodal content, with major upgrades to video, audio, and image generation aimed at creators and studios.

• Improved character and element consistency across scenes
  15-second video clips with better control and customizable multi-shot recording

• Audio now supports multiple character references, plus more languages and accents

• Image generation adds 4K output, continuous shooting, and more cinematic visuals

Try it Now → https://klingai.com/global/

2. Mistral drops Voxtral Transcribe 2

Mistral released two new speech models designed to deliver fast, accurate, and ultra-cheap transcription, without sending audio to the cloud. Built to run on laptops and smartphones, Voxtral is aimed squarely at regulated industries.

• Runs fully on-device with a 4B-parameter model, keeping sensitive audio local

• Batch transcription costs just $0.003/min and supports 13 major languages

• Realtime version hits ~200ms latency, ideal for live captions and voice agents

• Open-source (Apache 2.0) with downloadable weights on Hugging Face

Try it Now → https://mistral.ai/news/voxtral-transcribe-2

3. Anthropic Launches Claude Opus 4.6

Anthropic unveiled Claude Opus 4.6, its most capable model yet, built for sustained, real-world work across coding, research, and enterprise workflows. The company says it marks a shift from quick AI tasks to delegating serious professional work.

• Stronger at coding, planning, code review, and debugging across large codebases

• Excels at long-running tasks, research, and financial analysis

• Now ranks #1 on the Finance Agent benchmark

• Available via web, API, and all major cloud platforms

Try it Now → claude.ai

4. Roblox Launches 4D AI Creation Tools (Beta)

Roblox unveiled 4D generation, powered by its in-house Cube Foundation Model, allowing players and creators to generate not just 3D objects, but objects that behave, move, and function inside games automatically.

• Turn text prompts into fully functional objects (cars that drive, tools that work)

• Combines AI-generated meshes with behavior “schemas” for instant interactivity

• Removes the need for manual scripting or complex modeling

• Built directly into Roblox Studio and in-experience creation tools

Try it Now → https://www.roblox.com/

5. OpenAI Releases GPT-5.3-Codex

GPT-5.3-Codex is OpenAI’s newest coding-focused model, combining GPT-5.2-Codex’s programming strength with GPT-5.2’s broader reasoning and knowledge, while running significantly faster.

• Runs ~25% faster and uses fewer tokens than earlier Codex models

• Beats Anthropic’s Opus 4.6 by 12 points on Terminal-Bench 2.0

• Scores 64.7% on OSWorld, far ahead of GPT-5.2-Codex’s 38.2%

• Was used internally to find bugs, manage deployment, and evaluate its own training

Try it Now → https://openai.com/index/introducing-gpt-5-3-codex/

6. Moonshot AI Launches Kimi Agent Docs

Kimi Agent Docs is built to handle end-to-end document workflows, from long-form writing to conversion, review, and visual design, without manual formatting headaches.

• Generates Word or PDF documents up to 10,000 words with charts, formulas, and rich layouts

• Supports professional document types and bulk file creation in a single step

• Converts between Word, PDF, PPT, and Excel without losing structure or data

• Adds expert-style comments and revisions for academic, legal, and educational reviews

Try it Now → https://www.kimi.com/docs