Good morning. Google has a cheaper video model and a much bigger construction bill. Claude wants to explain your bank statement. Here are five developments worth understanding, and what, if anything, to do about them.
- Use: Google's new video model for a small, budgeted draft, not an automatic replacement for your designer.
- Watch: Claude's finance beta can read connected accounts, but important totals still need checking.
- Watch: Google's Wyoming data center is a $17 billion plan, not capacity you can buy today.
- Watch: Inkling 2 Small runs on Tinker; downloadable model files are promised later.
- Ignore, for general office work: GPT-6 Cyber starts with approved security customers.
The big story: Google lowers the price of a video draft
Google made Gemini Omni 1.2 Flash generally available on November 5. The model makes videos up to 15 seconds long, up from 10, and lets you ask for changes in a follow-up conversation. You can request a different background or lighting without starting a completely new brief.
Google quotes $0.07 per second for 720p video, a resolution commonly used for online drafts. That's 30% below its previous Omni 1.1 rate. At the quoted rate, one 15-second generation costs $1.05. Ten attempts cost $10.50 before any separately billed services. The useful budget is the cost of a clip you keep, not just the first one you generate.
Google also says the model follows reference images more closely and keeps people and objects more consistent between shots. Those are Google's claims. The company still identifies fast sports action and complex physics as areas needing improvement. A simple product scene is a more sensible first test than a football highlight reel.
Why it matters: Teams making product explainers or ad concepts have a clearer starting price for testing an idea before committing to a finished video.
Use Verdict — Use for low-stakes video drafts. The published rate is a starting price for a short test, not a finished video. If your team already needs a short concept clip, compare one simple scene with your current process. Set a spending limit first, then check the product's appearance, any lettering, and whether the result is actually usable. Keep a person responsible for approval. Recheck the quoted per-second price or the 15-second limit before you treat this rate as current.
Source: Google, November 5
Also useful: Claude can explain connected bank accounts
Anthropic's Personal Finance in Claude is rolling out in beta to US users on Pro, Max, and Ultra plans. It connects bank and card accounts through Plaid, a service that links financial apps to banks. Its access is read-only: Claude cannot move, manage, or invest your money.
The feature can add up spending, compare months, and identify charges worth another look. Anthropic says conversations inside the Personal Finance hub are not used to train its models. Chats outside that hub follow your usual privacy settings.
Why it matters: Someone who avoids spreadsheets may find it easier to ask why a month was more expensive, and then inspect the transactions behind the answer.
Watch Verdict — Watch for people who want help reviewing spending. Read-only access still exposes sensitive bank and card information. If you choose to use it, check important figures against your bank and ask which transactions support each total. Wait if you are uncomfortable connecting accounts. A clearer summary is not worth access you do not want to grant. Recheck if the beta stops being read-only or the listed US plans change.
Source: Anthropic
Behind the AI: Google's $17 billion Wyoming plan
Google announced its first Wyoming data center on November 5, describing a $17 billion long-term investment in Laramie County. A data center is a building filled with computers and the equipment that powers and cools them.
The energy plan includes a planned 702-megawatt fuel-cell development. Fuel cells turn fuel, in this case natural gas, into electricity through a chemical reaction. Google says the project includes technology to capture and store much of the resulting carbon emissions, and that the data center will use air cooling to reduce local water demands.
Why it matters: Building more AI capacity also means arranging electricity, cooling, land, and community support. An investment announcement does not tell customers when additional computing power will become available.
Watch Verdict — Watch for readers waiting on a construction or operating milestone. The dollar figure is an announced commitment, and the power figure describes the planned fuel-cell development, not verified computing capacity. Look for confirmed completion and service availability before connecting this project to any promise of faster or cheaper AI. Recheck when Google confirms the site is operating or the fuel-cell project is energized.
Sources: Google · Reuters, via Investing.com
Two quick updates
Inkling 2 Small: available to try, not yet to download
Thinking Machines introduced Inkling 2 Small on November 5. It accepts images and is available through Tinker, the company's platform for running and adapting models. The company intends to release the downloadable model files under an Apache 2.0 license after outside safety testing.
Why it matters: Trying a hosted model and running it yourself are different options.
Watch Verdict — Watch for people who need a model they can host themselves. The hosted trial is available, and the downloadable files are still a promise. Do not plan a switch until the files are actually posted. Recheck when the downloadable files and the stated Apache 2.0 license are posted.
Source: Thinking Machines
GPT-6 Cyber: a specialist release, not your next office chatbot
OpenAI introduced GPT-6 Cyber on November 5 for security research. Initial access is limited to selected approved Daybreak Red customers through its developer interface and Codex. OpenAI reports stronger results on its security tests. Those results do not establish whether it fits ordinary office tasks.
Why it matters: A major model announcement can be important without being something most readers need to adopt.
Ignore Verdict — Ignore for general office work. Initial access is limited to approved security customers, and the reported security results do not show an office-chat fit. Security teams should watch the access requirements and authorized testing rules. Everyone else can keep using the tools that already fit their job. Recheck if OpenAI offers it beyond those approved security customers.
Source: OpenAI
Your next move
Choose the one item tied to something you already need to do. If that's video, write down the scene and budget before opening the generator. If none fits your work, you're caught up. No new subscription required.
Sources & limits: Public sources checked November 8, 2026 (CST). Verdicts are editorial judgments for the audiences named, not hands-on test results. Plans and rollout details may change. Sponsors never buy a verdict.