Good Morning! Here's what I have for you in today's newsletter:
OpenAI introduces GPT-6 Astra, its most intelligent model yet
Google DeepMind unveils WeatherNext 3, its most accurate weather model
Microsoft launches MAI-Transcribe-2, 10x faster than GPT-Transcribe
Automate your research workflow with Fable 5.1 + GPT 5.6 orchestration
4 new AI tools worth trying today
AI AGENTS
OpenAI introduced GPT-6 Astra, its most intelligent and aligned model yet, setting a new state of the art for computer use, browsing, and software engineering. It's rolling out first to a limited set of organizations before reaching all ChatGPT and API users.
Astra leads on Agents' Last Exam, AutomationBench, and ScreenSpot Pro, benchmarks built specifically around real computer workflow tasks.
It's built to handle anything a person can do on a computer, clicking through apps, filling out forms, navigating websites, not just answering questions.
OpenAI calls it their most aligned model to date, with real improvements in understanding what a user actually wants.
Chaining clicks, form fills, and app switches into one finished task is a different bar than answering a question well. That's the difference between an agent that drafts an email and one that actually books the flight, fills out the form, and confirms the reservation.
AI MODELS
Google DeepMind introduced WeatherNext 3, their most advanced global weather model yet, using real-time satellite data to generate hourly high-resolution forecasts, precipitation predictions, and clean energy variables, now powering forecasts across Search, Gemini, Google Maps, and Google Earth Engine.
WeatherNext 3 trains directly on real-world satellite observations rather than relying on delayed prediction data, letting it generate hourly, high-resolution forecasts.
It produces precise precipitation forecasting and clean energy variables alongside standard weather data.
The model rolled out immediately across five different Google products, rather than launching as a standalone research release first.
A six-hour data lag sounds small until you need to know if it'll rain in the next hour. Skipping that lag by training on live observations gets forecasts closer to what's actually happening right now, not what was happening six hours ago.
AI TOOLS
Microsoft launched MAI-Transcribe-2, calling it the highest quality, cheapest transcription at the fastest speed, running 10 times faster than GPT-Transcribe and now available on Microsoft Foundry.
The model claims to lead on three separate axes at once, quality, price, and speed, rather than trading one for another.
A 10x speed improvement over GPT-Transcribe specifically targets latency-sensitive transcription workloads.
Availability through Microsoft Foundry puts this directly in front of enterprise developers already building on Microsoft's platform.
Transcription tools usually force a tradeoff between speed, cost, and accuracy. Claiming all three at once, at 10x the speed of a direct competitor, is a big claim that's easy to test yourself once it's actually live.
HOW TO AI
Research and content work has a genuine bottleneck, one person doing the searching, drafting, and checking, all in sequence, all with the same model.
A free plugin called fable-advisor splits that up.
Fable 5.1 runs the session and checks the final result. The actual work goes to whichever model fits, including GPT-5.6 through OpenAI's Codex CLI.
Step 1: Install fable-advisor
claude plugin marketplace add DannyMac180/fable-advisor
claude plugin install fable-advisor@fable-advisorStart your session with Fable in charge:
/model fableNo Fable access? Swap model: fable for model: opus in the plugin's agents/fable-advisor.md file.
Step 2: Connect GPT-5.6 to do the actual work
Needs OpenAI's Codex CLI, logged in with your own ChatGPT account.
npm i -g @openai/codex
codex loginOnce connected, fable-advisor hands work to two GPT-5.6 lanes, one for routine tasks, one for higher-complexity ones.
Step 3: Give it a research task
Install last30days, which researches a topic across Reddit, X, YouTube, and the web from the last 30 days.
/plugin marketplace add mvanhorn/last30days-skillUse /last30days to research what the AI community is actually saying about Claude Fable 5.1 versus GPT-5.6 for agentic tasks, then combine that with official documentation to give me a clean comparison
Step 4: Understand the review step
Once the work finishes, Fable reviews the result against what was actually asked, not just skimming for obvious errors.
Step 5: Turn the workflow into a skill
Turn this exact research-and-review workflow into a reusable skill I can call with one command
P.S. You can access all the AI trainings (including the full version of this one), prompts and workflows if you upgrade.
Anthropic added /ant apply to the ant CLI, letting developers declare Claude Managed Agent environments, agents, skills, memory stores, and deployments as files in a repository, then keep the API's resources synced with them.
Perplexity launched its API in Stripe Projects, letting anyone provision a Perplexity API project, key, and prepaid credits straight from the terminal or a coding agent using the Stripe Projects CLI.
Browser Use introduced payment support through Link, letting a web agent make purchases with a connected credit card without ever seeing the real card details, using a single-use card that dies after one purchase.
🖥️ Hermes Desktop: now sets up local models in one click, automatically reading your hardware, picking the best model, and configuring the runtime.
🔗 Lovable: now available from Linear, turning a spec sitting in a Linear issue into a live, hosted app with one call.
📊 Adaption Labs AI describe the dataset you need and get a structured, training-ready dataset back, no existing data required.
🌍 Atlas: the first multimodal world model with pixel-perfect camera control that reconstructs scenes in 3D, currently early access only.
THAT’S IT FOR TODAY
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
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial