Hi tech & AI enthusiasts, this week's highlights:
OpenAI Just Launched GPT-6 Astra.
Meta Surprises With Muse Spark 1.3
Google Admits It's Below the Frontier, Then Ships an Upgrade
The Lab: Turn an Illustration into an Interactive 3D Scene with AI
AI Image of the Week: A Japanese house illustration for 3D modelling
Let's get into it.
▲ AI MODELS
OpenAI released GPT-6 Astra, calling it the "most intelligent and aligned model in the world." It sets new highs across science, math, coding, cybersecurity + computer use. And it’s my go to model for highly complex tasks.
What you need to know
Record scores, middling rank: Astra topped several individual tests, but sits at 61 on Artificial Analysis's Intelligence Index.
Higher price, fewer steps: It costs roughly 2.5x more than GPT-5.6 Sol, but completes complex tasks in fewer attempts.
The AGI question: OpenAI's president said "For me personally, I do think we're there." It’s a personal view, not a company position, but it is the strongest statement from anyone at a major lab.
Available now on paid plans: Paid ChatGPT users and API customers can access it today, with an Astra Pro tier rolling out for Pro subscribers.
Key Idea
▲ Why it matters: Benchmarks are one thing. Everyday tasks is what is more important. Astra is the best AI model I’ve tested. Coding it doesn’t need much hand holding. For general work. It’s smarter and uses the computer faster. If you haven’t already I’d try using it within your workflows ideally within ChatGPT Work or Codex.
▲ OPEN MODELS
Meta is back in the game. Muse Spark 1.3 puts the company close to frontier from Anthropic and OpenAI but at a fraction of the cost.
What you need to know
Near the top of the table: Spark 1.3 (Max) scores 62 on the Intelligence Index, trailing only Claude Fable 5.1 and Opus 5.
Free right now: You can use it today on OpenCode for free. (I use it within Hermes)
Cheap when you do pay: API pricing sits well below comparable offerings from OpenAI and Anthropic.
More coming: Zuckerberg said "next up" is a larger model codenamed Watermelon, plus releasing the Spark weights as open source.
Key Idea
▲ Why it matters: Zuckerberg stated "frontier performance almost too cheap to meter." If you’re running AI-heavy workflows or building tools, Meta just made that significantly cheaper. Companies are looking to cut AI costs and if Watermelon is even better. There will be pressure on everyone else to keep costs down.
▲ AI MODELS
Google doesn’t have an AI model competing with OpenAI’s Astra or Anthropic’s Fable. This is a problem that they know they are facing. But what they do have is an incredible capable model for everyday use. Gemini 3.8 Flash has just been released and it is better than most give it credit for.
What you need to know
Same price, more capable: 3.8 Flash keeps the same pricing as 3.7, with gains in coding, reasoning + agentic tasks. It scores 59 on the Intelligence Index.
Google knows they’re behind: DeepMind's Koray Kavukcuoglu admitted Gemini sits "a little below the frontier" and said "there's nothing other than being at the frontier that's important for us."
The Pro model is MIA: Google's larger Pro tier appears to be on hold until Gemini 4. That leaves Flash carrying the full weight of Google’s lineup.
Key Idea
▲ Why it matters: For daily tasks and fast automated workflows, 3.8 Flash is a solid, affordable option. I’m using it for some automations & apps due to it’s price and performance. For your most complex work. I’d avoid. But if you’re wanting to build out ideas or save on costs Flash will handle that well. I wouldn’t discount Google yet while we wait on their Pro model.
▲ EVEN MORE NEWS
Other big ideas from the past week.
▲ WorldLabs Releases Atlas: Every time I see a WorldLabs release I’m blown away. This time they’re created the first multimodal world model. With just a couple of cameras they can generate a full 3D spatial scene. (it’s easier to just see the video above.)
▲ Google's Top Music Model Comes to Gemini: Lyria 3.5 is now live in the Gemini app, AI Studio + the API. Pick genres, choose vocal or instrumental styles, and use templates for anything from background music to custom birthday tracks.
▲ Alibaba's Qwen Upgrades Its Top Model: Qwen3.8-Max-0902 brings a major boost in coding + collaborative agent work. Stronger at complex enterprise tasks and long-running workflows. A serious option if you need heavy lifting at a lower price point.
The Lab
▲ AI EDUCATION
This week you’ll learn how AI handles 3D designs.
You will take a Japanese townhouse illustration and turn it into a fully interactive 3D scene using Gemini 3.8 Flash and Muse Spark 1.3.
Both results are live at example.metalabs.global for you to compare.
Work that used to take a 3D team weeks can now be done in a few hours.
What you'll learn:
How to generate clean architectural reference art with ChatGPT
The exact build brief that turns 2D geometry into Three.js code
How to steer and refine the scene across multiple iterations
How Gemini 3.8 Flash and Muse Spark 1.3 compare on 3D code generation
How to deploy the single-file result live in seconds
🔬 Interested in learning about AI for business? Join The Lab waitlist.
▲ AI IMAGE SHOWCASE
Japanese House Illustration for 3D Modelling
This week's image matches our 3D building tutorial.
Prompt:
A warm digital illustration of a traditional Japanese residential house. Two storeys, dark wooden frame, pale plaster walls, sloped clay tile roof with curved eaves. A small covered entrance with a noren curtain. Paper lanterns glowing soft amber, hanging beside the doorway. A compact front garden with a stone path, trimmed shrubs, and a single maple tree with red-orange leaves. Evening light, warm golden tones, soft shadows. Detailed architectural concept art style, clean lines, slightly textured paper background. No people. Front-facing composition with breathing room on all sides.Tool: Gemini 3.1 Flash Image
Image Usage Suggestion: Architectural mood boards, real estate marketing concepts, construction and renovation before/after comparisons, travel blog headers, hospitality brand social content, or as a reference image for the 3D modelling process covered in this week's tutorial.
💌 Reply to this email with your own version and I'll feature it in a future newsletter.
▲ SHARE YOUR THOUGHTS
Help shape the content you see here by giving feedback
Have specific feedback or want to get in touch? Reply to this email and I’ll get back to you.
Know someone who’d love this newsletter? Forward it to a friend and have them sign up here.
Thanks for reading, until next time.
Stay curious,
Matt Lok, Editor
Brought to you by Metalabs. Digital marketing consulting specialising in AI. Carousels made with SlideDrift.