Good Morning! Here's what I have for you in today's newsletter:

  • OpenAI releases GPT-6.2 Astra and Sol, with Astra API prices 20% lower

  • Google opens Tasks in Gemini to free users in six countries

  • Microsoft launches seven MAI models, led by MAI Code 2 for Copilot

  • Make AI sound human with a Claude skill built from your writing

  • 4 new AI tools worth trying today

AI MODELS

OpenAI released GPT-6.2 Astra for complex projects and GPT-6.2 Sol as a lower-cost option for everyday coding and writing. Both are available through the API and in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.

  • Astra writes more directly with less filler and follows professional templates more closely. OpenAI says it improves financial models, presentations, spreadsheets, legal research and software engineering work.

  • Astra’s API rates fall 20% to $8 per million input tokens and $40 per million output tokens. Cached input drops from $1 to $0.40 per million tokens. Sol stays at $2 for input and $10 for output.

  • OpenAI says Sol comes close to Astra on FrontierCode 2.0 at roughly one-fifth of the cost. Pro 500 subscribers get Ultrafast on both models, while Pro 200 subscribers get it on Sol.

Every report, spreadsheet and code fix you hand to a model shows up on your API bill. Astra takes on the bigger projects at a lower price than before, while Sol handles routine coding and writing for a fraction of that. If you’re on Pro 200, you can now try Ultrafast with Sol for quicker replies.

AI AGENTS

Google opened Tasks in Gemini to free personal accounts for users 18 and older in the US, Japan, South Korea, Canada, Australia, and India. The mode runs on Gemini 4 Hydrogen and handles errands, bills, and other to-dos.

  • In the US, Gemini can call several businesses at once to compare quotes or find an appointment. It asks before dialing and provides a transcript afterward.

  • US users can also check out through Google Pay using the Agent Payments Protocol. Gemini asks before buying. Tasks can also be sent by text or email without opening the app.

  • Outside the six free markets, Tasks requires Google AI Pro or Ultra where supported. It remains unavailable in the EEA, the UK, Switzerland, and Nigeria, including on paid plans.

Moving Tasks onto the free tier widens access to agents that can act outside a chat. The US rollout also shows where Google is drawing the line for now: Gemini can make calls and complete purchases, but people approve those actions first. That approval step becomes more consequential as consumer agents interact with businesses and payment networks at scale.

AI CODING

Microsoft launched seven MAI models at Ignite, including MAI Code 2 and MAI Code 2 Flash for GitHub Copilot and VS Code. Microsoft says Code 2 reaches 96% of Claude Opus 5.5’s coding performance at about 5% of the price.

  • Code 2 Flash handles quick edits and routine fixes on Copilot Free, Student, and paid plans. Microsoft says it matches GPT-6 Luna on everyday tasks at 65% of the price.

  • MAI Cyber 2 Flash is limited to verified security defenders. Microsoft reports 92.3% on CyberGym-E2E, compared with GPT-6 Astra’s 91.5%, at less than 2% of the cost per task.

  • The other releases are MAI Image 2.7 in Microsoft Foundry, MAI Realtime 1 for live voice starting at $0.01 per minute through MAI Labs, Dragon Copilot Transcribe, and Decision 1.1 for routing requests to models.

Microsoft can put its own coding models in front of developers through Copilot and VS Code. If its price and performance claims hold up in day-to-day work, it can route more routine requests to its own models and rely less on outside providers. That gives Microsoft more control over Copilot’s costs and puts pressure on rivals to compete on price as well as performance.

HOW TO AI

Turn five pieces of your writing into a Claude skill you call with /my-voice.

Give Claude examples of how you write, and it can use them to build a voice card, list the habits you want it to avoid, and save both as a skill. This walkthrough uses Opus 5.5 on a paid Claude plan.

Step 1: Pick your samples

Start a chat on Opus 5.5. Paste five pieces you wrote without AI help and label each by format, such as email or post. Aim for 300 words or more per sample where practical. Remove client names, private details, and other people’s messages before pasting. The skill keeps the samples in full.

Step 2: Map your voice

Ask Claude for a voice card covering sentence length, openings, words you use and avoid, punctuation, tone, and endings, with a quote from your samples to support each point. Review the claims about words you never use. Five samples can show tendencies, and you can confirm which words you deliberately avoid.

Step 3: List the AI tells

In the same chat, ask for a never list of stock phrases, punctuation, and sentence patterns that would sound out of place in your writing. Remove anything you’d say yourself and add the habits you routinely edit out.

Step 4: Turn it into a skill

Ask Claude to save the voice card, never list, and samples as a skill called my-voice. Review the description on the Save plugin card, open Content to check what’s included, then click Save. Look for My voice marked Added under Settings, then Plugins.

Step 5: Test it blind

Use /my-voice in a new chat to draft 200 words on a topic you’ve written about. Fill in any placeholders. Shuffle the draft with two pieces of your own writing in a similar format, then ask a fresh Claude chat which one sounds AI-written and which lines gave it that impression. If a quoted line sounds unlike you, turn that feedback into a specific rule for the skill.

P.S. Become an AI Operator here.

Anthropic added an experimental Code Mode to Claude Code. Claude can write a script that chains several tools together and keep replying while a long command runs. The feature is opt-in.

Google brought Android app building to AI Studio. Gemini can build a full-stack app that you install from a personal Google Play library. Play installs are US-only for now, and web apps can use custom domains on paid Cloud projects.

Perplexity added a Multiagent effort level to Computer for Max subscribers on the web. A lead agent divides a task among up to 16 agents running different models, and they cross-check one another’s work.

🖼️ Kling: Launched Image 4.0 with native 8K output, support for up to 2,000 characters of text per image, and batches of up to 10 images with a consistent look.

🌐 GitHub Copilot: Added a Chrome and Edge extension that can fill forms, pull information from pages, and test web apps in a signed-in browser.

🎬 HeyGen: Released Avatar 6, which builds a native 2K avatar from 30 seconds of video and a photo, with gestures matched to the script.

👕 Ideogram: Launched Print on Demand Studio, bringing product research, design, print preparation, and mockups into one workspace.

Which image is real?

Login or Subscribe to participate

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

✓ $500+ worth of AI tools