In partnership with
🧠 THAT ONE AI - This Week’s Signal
Here’s what’s shaping AI right now - without the noise:
⚔️ Astra vs Fable 5.1: Which Tab Do You Open?
💥 GPT-6 Astra reached ChatGPT, with fewer messages per window
💰 The top 1% of customers give OpenAI and Anthropic 80% of revenue
🧠 How to make each Astra message count
🧰 Tools worth testing
⚔️ Astra vs Fable 5.1: Which Tab Do You Open?
Both labs shipped their best model in the same week. Most people now pay for both and guess which one to use. Here is the short answer.
📊 The scoreboard
GPT-6 Astra wins on speed and price. It runs about 2 times faster than Claude Fable 5.1. It also costs about 2 times less. It leads on most public benchmarks, and computer use is its largest lead of all.
One number goes the other way. On Artificial Analysis's combined Intelligence Index, which mixes many tests into 1 score, Fable 5.1 still finishes above Astra. So the phrase "best model" depends on which test you trust.
🖥️ What Astra does better
Astra operates a computer well. It works inside spreadsheets, PDF files, slide decks, images, and video. Give it a clear finish line, and it does the work itself.
Astra also asks more clarifying questions than older models. This helps at the start of a new project. It can also interrupt a task that you wanted it to finish alone. Name the 1 decision that needs your input, and let Astra decide the rest.
📚 What Fable 5.1 does better
Fable 5.1 goes straight to the result. It gives fewer progress updates than earlier Claude models. Ask for a short status line if you want to follow the work.
It also reads dense documents well. Charts and complex tables inside PDF files are its strongest area. Its memory ends in June 2026, so tell it to search the web first for recent facts. Ask it to put copied text in quotation marks, because it does not always mark quotations itself.
🎯 How to choose in 5 seconds
The task runs on your computer, across apps and files → Astra
The task is long analysis of dense documents → Fable 5.1
You want speed and low cost → Astra
You want careful writing that you will publish → Fable 5.1
⚙️ Settings that change the result
In ChatGPT, use Work mode or Codex mode. Then set effort to Medium. Medium gives the best balance of quality and price, and it answers fast.
In Claude, open the model selector and pick Fable 5.1. Start serious work at High effort. Switch to Medium after the first 2 turns. Long chats burn more tokens, so start a new chat when the thread stops moving forward.
💬 2 prompts to keep
For Astra:
Deliver [exact output]. Success means [criteria]. Use the attached files as [evidence / style reference / template]. Ask me only about [the one decision you need from me]. Complete everything else, then list your assumptions at the end.
For Fable 5.1:
Use ASD-STE100. I want [goal], and success means [criteria]. Search the web first and cite your sources. Put copied text in quotation marks. Ask me clarifying questions until you have enough context to start.
Read the official guidance: OpenAI's prompting notes for Astra and Anthropic's prompting notes for Fable 5.1.
The bigger signal: 👉 The 2 models now win at different jobs, so your real skill is picking the right tab and the right effort level before you type anything.
200+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 200+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
200+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
💰 The Top 1% of Customers Give OpenAI and Anthropic 80% of Revenue
New data from Ramp shows heavy revenue concentration at both labs. The top 1% of customers produce 80% of enterprise revenue at OpenAI and at Anthropic. This pattern holds even as more companies start to pay for AI.
Ramp lead economist Ara Kharazian says he sees no other software category with this level of concentration risk. Usage-based pricing explains it. Spend follows how much work the software does, not how many people log in.
At Anthropic, 2 coding tools show the effect clearly. Cursor and GitHub Copilot together produced about $1.2 billion of Anthropic's $5 billion revenue last year. OpenAI has not said what share of its revenue any single customer represents.
The bigger signal: 👉 If you build on these APIs, your provider depends on a few very large accounts, and its pricing can move to protect them.
💥 GPT-6 Astra Reached ChatGPT, With Fewer Messages Per Window
OpenAI released GPT-6 Astra to ChatGPT users on Pro, Enterprise, and Business Premium plans. Access came first through ChatGPT Work and Codex. Plus and Business users get the model over the next few days. Free and Go users do not get it at all.
Astra also reached developers through the OpenAI API, Microsoft Azure, and AWS Bedrock. Pro, Business, and Enterprise plans include GPT-6 Astra Pro, the stronger version. All usage counts against the quota of your existing plan.
The message limits changed too. Standard Astra gives about half the messages per 5-hour window that GPT-5.6 Sol gave. Plus users get about 5 to 45 messages. Sol gave them 10 to 100.
The bigger signal: 👉 Astra costs you more per question than Sol did, so decide which tasks really need the newest model before you spend a turn.
🧠 That One AI Tip: Make Each Astra Message Count
Astra gives you fewer messages per window than Sol did. Treat each message as a unit that you must not waste. Use this method.
Write the brief before you open the chat. Decide the exact output that you want. Name the format, the length, and the reader. A vague first message wastes a full turn.
Put all context in the first message. Attach the files, the data, and the limits together. Do not send them one at a time. Each follow-up message costs the same as the first one.
Draft with a cheaper model. Use GPT-5.6 Sol or another model for early drafts and simple edits. Send only the hard reasoning work to Astra.
Keep a reusable brief file. Save your best prompts in a plain text file. Paste the brief, then change only the task line. This keeps your setup cost near zero.
Task: [what you want]
Output: [format, length]
Reader: [who reads it]
Limits: [rules, tone, things to avoid]
Context: [files, data, background]The bigger signal: 👉 Message limits change how you work, so a repeatable prompt structure is now worth more than a clever prompt.
That One AI 🧰 TOOLBOX
A few tools quietly worth exploring:
🌎 Atlas → Builds a 3D world from text, an image, or a video, and lets you move inside it.
🎨 Moda → Creates on-brand visual assets on a canvas that you control and can edit.
🎬 Higgsfield Genjutsu → Swaps characters, clothes, and objects in your video, and keeps the original motion.
🗓️ Clockwork → Schedules AI coding agents on a calendar and runs them in sandboxed git worktrees.
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
This guide distills 10 AI strategies from industry leaders that are transforming marketing.
Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI
Learn how AI-driven emails achieved 94% higher conversion rates
Discover 7 ways to enhance your marketing strategy with AI.
🔚 EXIT NODE
The top labs showed 2 different kinds of progress this week. Claude finished a 358-year-old proof with very little human help. OpenAI shipped its strongest model, then cut how often most people can use it. Models get stronger. Access to them gets tighter.
Before you go - what do you want us to cover next? Reply to this email and let us know.
See you next issue.