AI Weekly Recap (Week 39)

Plus: This week’s top AI news, breakthroughs, and game-changing updates

Author

Alvaro Cintas
September 28, 2025

Good morning, AI enthusiast. AI just had a wild week. OpenAI launched ChatGPT Pulse for personalized morning briefs, while Alibaba unveiled Qwen-Max, a 1T-parameter model for reasoning, coding, and multimodal tasks.

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

OpenAI is rolling out ChatGPT Pulse, a feature that generates personalized reports while users sleep. Pulse delivers 5–10 briefs each morning, summarizing news, events, or personalized tasks, and is currently available to Pro subscribers.

→ Generates proactive AI reports with text & images
→ Works with Gmail, Google Calendar, and other ChatGPT Connectors
→ Uses memory and context from previous chats for personalization
→ Users can request new reports or provide feedback for refinement

🧰 Who is this useful for:

  • Busy professionals who want a morning digest without doomscrolling

  • Creators, founders & execs needing daily context from emails + calendars

  • AI early adopters testing assistant-like features vs. chatbots

  • Pro subscribers ($200/mo) who want more proactive AI support

Suno has released Model v5, its upgraded AI music tool, with major improvements in sound quality, but critics say it still feels “soulless.”

→ Cleaner mixes: better separation between instruments, fewer artifacts
→ Smarter at recreating effects (e.g. stereo delays, synth textures)
→ Vocals remain hyper-polished, always on pitch, reverb-heavy, harmony-laden
→ Claims improved genre understanding, though early testing raises doubts

🧰 Who is this useful for:

  • Producers curious about AI-assisted track building

  • Content creators needing quick, polished backing tracks

  • Musicians exploring AI as a tool for demos or experiments

  • Researchers tracking the limits of creativity in AI art

Try it now → https://suno.com/home

Perplexity AI has launched Email Assistant, an inbox-management tool for Max subscribers that works with Gmail and Outlook. It acts like a personal assistant, drafting replies, organizing emails, and scheduling meetings, with enterprise-grade privacy protections.

→ Drafts replies + proposes meeting times based on calendar
→ Summarizes conversations & prioritizes important emails
→ Smart labels: urgent, action-needed, completed tasks
→ SOC 2 + GDPR compliant, with no training on personal data

🧰 Who is this useful for:

  • Executives & professionals buried in email overload

  • Teams needing AI-powered scheduling + follow-ups

  • Enterprises requiring secure AI tools for sensitive comms

  • Power users testing AI assistants across workflows

Alibaba’s Qwen team released six major model variants this week, targeting different AI tasks from reasoning to translation and safety.

→ Qwen-Max (1T params): Near-frontier coding + agentic abilities; Heavy version aces math reasoning benchmarks
→ Qwen-Omni: Multimodal (text, images, audio, video) with speech support in 19 languages (understanding) & 10 (generation)
→ Qwen-VL: Top open-source visual model, outperforming even leading closed models on benchmarks
→ LiveTranslate-Flash: Real-time speech interpretation
→ Guard models: Advanced safety + moderation capabilities
→ Upgraded Coders: Improved reasoning + developer support

🧰 Who is this useful for:

  • Developers seeking cutting-edge open models for apps & agents

  • Enterprises building safe, multilingual AI workflows

  • Researchers benchmarking frontier reasoning & vision model

  • Global teams needing real-time AI translation + moderation

Factory’s Droids have become the top-ranked agents on Terminal-Bench, outperforming Claude Code, Codex CLI, and others, and are now expanding beyond software development. The company also raised $50M from NEA, Sequoia, J.P. Morgan, Nvidia, and top tech leaders.

→ #1 on Terminal-Bench, the toughest general software dev benchmark
→ Works with any LLM, any IDE, any interface (CLI, IDE, Slack, Linear, Browser)
→ Supports local + remote environments, plus “headless mode” for automation
→ Extensible with MCPs for planning, iteration, and building custom workflows

🧰 Who is this useful for:

  • Developers tired of being locked into single IDEs or platforms

  • Teams needing flexible agents across coding + collaboration tools

  • Enterprises modernizing legacy codebases and workflows

  • Power users automating custom pipelines with headless mode

Google has launched Mixboard, a new AI app that creates mood boards directly from text prompts, no saved Pins needed. The service is now live in the U.S. as a public beta via Google Labs.

→ Generate mood boards instantly with AI or use pre-populated templates
→ Powered by Nano Banana, Google’s advanced image-editing model
→ Edit, combine, or regenerate images to spark new ideas
→ Supports brainstorming with both images + text

🧰 Who is this useful for:

  • Creators exploring home décor, event design, or DIY projects

  • Brands and teams brainstorming campaign or design ideas

  • Gen Z users who loved Pinterest Shuffles for TikTok-style content

  • Anyone wanting quick, aesthetic inspiration without curating Pinse

Alibaba has released Wan2.5, its most advanced multimodal visual generation model yet. It introduces a unified architecture for text, image, video, and audio, with cinematic-quality video and precise image editing.

→ Native multimodal architecture for text, images, video & audio
→ Joint multimodal training for stronger audio-visual sync
→ RLHF alignment for more realistic images & video dynamics
→ Generates 1080p, 10s cinematic-quality videos with audio
→ Pixel-level image editing with creative typography, fusion & color swaps

🧰 Who is this useful for:

  • Filmmakers & studios exploring AI-driven production

  • Brands creating cinematic ads or product visual

  • Designers needing creative editing + pro-grade charts

  • Researchers tracking multimodal “world model” development

Try it Now → https://wan.video/

Meta Platforms launched Vibes, a new feed of AI-generated short-form videos, aimed at accelerating its AI efforts and giving users creative tools to make, remix, or enhance video content.

→ Create videos from scratch or remix existing content
→ Add new visuals or layer in music
→ Upload directly to Vibes or cross-post to Instagram & Facebook stories/reels
→ Available on the Meta AI app and Meta AI website

🧰 Who is this useful for:

  • Content creators experimenting with AI-enhanced video

  • Social media marketers boosting engagement on Instagram/Facebook

  • Users looking to remix and personalize short-form videos

  • Anyone curious about AI tools for creative expression

That's it! See you tomorrow

- Dr. Alvaro Cintas