Google and Apple both shipped AI with the catch in the fine print: a price that doubles Nov. 30 and a daily cap with a fee to come. Read the footnotes.
Google released Gemini 4 Argon on Oct. 13 at $2 per million input tokens and $10 per million output, rising to $4 and $20 after Nov. 30. The intro price is a seven-week test window, so test now and budget at full price.
Why it matters If you build on Claude or GPT, Google just paid for your next eval. Run your real prompts against Argon before Nov. 30, and price any switch at $4 and $20, because that is the December bill.
SRCGoogle
Apple opened pre-orders for a $299 Apple TV 4K and the $299 HomeView, both built around a Siri AI beta and available Nov. 13. Apple's fine print caps server features daily and says more access will cost a fee.
Why it matters Apple just confirmed its assistant is going metered. If you build smart home gear or apps that lean on Siri, design for a customer who has hit today's limit, and if you are in the EU, none of it is coming yet.
SRCApple
Google says you can now build an Android app in AI Studio and run it on your phone through Google Play with a standard Google account, no developer account needed. Vibe coding just got a home screen icon.
Why it matters The last step between describing an app and using it is gone for anyone 18 or older in the launch markets. Build the small tool you keep meaning to, and keep it away from your passwords until you know what it does.
SRCGoogle
Prosecutors charged OpenAI engineer Abdulrahman Alfozan with planting a recording device in an executive conference room, citing prompts about building one from a linked chatbot account. Your chat history is a record a court can read.
Why it matters Two lessons for anyone who uses AI at work. What you type into a chatbot can end up in a criminal complaint, and the cheapest attack on an AI company is an insider, not a hacker.
An open letter signed by 415 researchers, including Hinton, Bengio and Russell, asks AI labs to airgap cyber-capable models by default and report containment failures. The labs will resist, but your agents need the same discipline.
Why it matters If you hand an agent a browser, a shell or API keys, you are running a small version of the risk this letter describes. Sandbox it, limit its network access and log what it does this week.
Decide whether to switch an AI workload before a price change
› I run an AI workload and want to decide whether to test or switch to a new model while it is on an introductory price. Act as a blunt cost and evaluation advisor.
My workload: [what the model does, e.g. summarizing support tickets]
Current model and price: [model, $ per 1M input tokens, $ per 1M output tokens]
New model and prices: [intro input/output price] until [date], then [full input/output price]
Monthly volume: [input tokens], [output tokens], share of input that is cached or repeated: [%]
What I care about most: [quality, speed, cost, reliability]
Do this:
1. Calculate my monthly bill on the current model, the new model at the intro price, and the new model at the full price. Show the math in a small table.
2. Tell me the break-even: how much better the new model must be (fewer retries, shorter outputs, fewer human fixes) to be worth it at the full price.
3. Write a test plan I can run before the intro price ends: 20 real inputs from my own work, how to score them side by side, and what result means switch, stay, or split traffic.
4. List three risks of switching (prompt changes, output format drift, rate limits) and how to check each.
Ask me for any number I left out before you calculate. Do not assume.Works in any AI chat. In Magai, run it against several models at once.
Each square is an edition. Click any day to read the five that mattered then.
Every term our stories use, defined in plain English and linked wherever it appears.
The glossary →