Hi tech & AI enthusiasts, this week's highlights:
Claude Opus 5 Lands at Half the Price
FLUX 3 Makes Video, Images and Sound From One Model
Gemini 3.6 Flash Cuts the Cost of Running AI
The Lab: How to Create a Claude Skill by Recording Your Screen
AI Image of the Week: Robots on the factory floor
Let's get into it.
▲ AI MODELS
Anthropic released Claude Opus 5 across the Claude apps and the API. It gets close to the intelligence of their most expensive model, at half the cost, and is priced exactly the same as the version it replaces.
What you need to know
Their best model yet for actual work: It now leads the industry tests that measure real business tasks rather than coding puzzles. On one automation test it finished every job it was given, where previous Claude models failed outright.
It finds its own way around problems: Given a technical drawing it could not open, it built itself a tool to read the file and rebuild the part. No other model managed it in five attempts.
Less likely to make things up: Anthropic's own testing shows this is their least deceptive model to date, and the hardest to trick into doing something it should not.
Key Idea
▲ Why it matters: Anthropic keeps running the same play: take the intelligence from their most expensive tier and halve what it costs. Opus 5 costs exactly what the old version did, so if you already pay for Claude you got a better model this week for nothing. If their top model was out of your budget, this is now the sensible option.
▲ AI TOOLS
German AI lab Black Forest Labs opened early access to FLUX 3, a single model that makes video, images and matching audio. A version of it is already controlling robot arms on Audi's production lines.
What you need to know
20-second clips with sound built in: Dialogue in multiple languages, readable text on screen, and characters that stay consistent from shot to shot. Feed it words, a photo, or existing footage.
It beat every rival it was tested against: In blind comparisons against the main video tools, testers picked FLUX 3's output most of the time.
The same model is running robots: A variant is being tested on Audi's factory floor, guiding machinery rather than making video.
Key Idea
▲ Why it matters: One model that handles video, images and sound together means fewer tools and less stitching for anyone producing content. The robotics part is the bigger story though. The same technology that makes a video look convincing is now good enough to guide a machine on a factory floor, and that is a far larger market than marketing.
▲ AI MODELS
Google released Gemini 3.6 Flash, alongside a lighter version for high-volume work. It performs better than the model it replaces and costs less to run.
What you need to know
Cheaper and less wasteful at once: It costs less per job than the version it replaces, and gets there in fewer steps. Both halves of the bill come down together, which rarely happens.
Noticeably better at the boring work: Legal firm Harvey and research platform Hebbia both report it handles document reading, chart analysis and report drafting better than before. Coding accuracy improved by roughly a third.
A lighter option for volume: The stripped-back version is built for jobs you run thousands of times, like searching and sorting documents, at a fraction of the cost.
Key Idea
▲ Why it matters: Cost is what separates an AI tool you demo once from one you run every day. When the price drops far enough, jobs you would never have paid a person to do become worth automating. That is the real shift here. Google also confirmed work has started on Gemini 4.
▲ EVEN MORE NEWS
Other big ideas from the past week.
▲ You Can Now Talk to ChatGPT on Your Desktop: OpenAI brought voice to the Mac and Windows apps for paid plans, so you can speak to it rather than type.
▲ An AI Designed a Working Computer Chip in 48 Hours: Chinese lab Moonshot released Kimi K3, the largest model anyone can download and run for free. In one test it designed a working chip on its own over a weekend.
▲ Meta AI Starts Doing Things, Not Just Answering: Meta AI can now connect to your email and calendar, build slides, run a daily briefing, and follow through on recurring tasks without being asked again each time.
The Lab
▲ AI EDUCATION
Claude can now watch you do a task once, then turn it into a reusable skill. No prompt writing, no configuration. You record your screen doing the job the way you normally would, and Claude writes the skill for you.
What you'll learn:
How to set up screen recording in Claude for Mac
How to record a repeatable business task, using my own weekly newsletter republishing job as the worked example
How to narrate while you work so Claude captures the why, not just the clicks
How to review, test, and refine the skill Claude proposes
🔬 Interested in learning about AI for business? Join The Lab waitlist.
▲ AI IMAGE SHOWCASE
Create this type of image with ChatGPT Images 2
This week's FLUX 3 story showed a visual AI model being used to control robots on Audi's production lines. Construction is heading the same way through off-site and modular manufacturing, so this week's image puts a robot and a human on the same wall panel.
A wide cinematic photograph inside a bright modular construction factory. A robotic arm and a human worker in a hi-vis vest work side by side assembling a large timber-frame wall panel on a jig. Shafts of daylight fall from high clerestory windows, fine sawdust suspended in the air. Muted industrial palette of pale timber, steel grey and safety orange. Shallow depth of field, photorealistic, shot on a 35mm lens at f/2.8.Image Usage Suggestion: Modern methods of construction content, off-site and modular manufacturing marketing, automation and future-of-work articles, recruitment campaigns for tech-forward contractors, tender submissions that need to signal innovation, or LinkedIn posts about where the industry is heading.
💌 Reply to this email with your AI image generation or suggestion of what you'd like to see 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.