Your customers just got a new idea of what an answer looks like. Since yesterday, ChatGPT replies with buttons, charts, forms and little working calculators instead of walls of text. Paying users got it on October 7. Free users get it today.
That's the lead story, and it's bigger than another model number. When the most-used AI app on earth starts answering like a tiny app, people bring that expectation to everyone else they talk to. Including your DMs. Let's go through the day.
ChatGPT stops answering in paragraphs
OpenAI is rolling out GPT-6 to everyone in ChatGPT, along with something it calls Intelligent UI. Ask a question and the answer can arrive as a chart, a side-by-side comparison, a checklist, a map, or a small tool. OpenAI's own examples include a recipe widget that rescales ingredients for your guest count and a bill splitter. Follow-up questions can show up as tap-to-choose options.
The split is simple. Plus, Pro, Business and Enterprise users get GPT-6 Sol. Free and Go users get GPT-6 Luna. Nobody has to flip a setting. OpenAI also says the model can start answering while it's still thinking, which it claims cuts wait times by 44 percent. For now this lives in the regular chat tab only.
Is it new? Not entirely. Google gave Gemini a similar visual answer style back in May. But ChatGPT is where hundreds of millions of people ask "which one should I buy?", so this is where the habit will form.
Why it matters for you: your shopper is about to spend the evening getting instant, tidy, tappable answers from an AI, then open your page, ask "price?" and wait two hours for a reply. The gap isn't about fancy buttons. It's about speed and clarity. A fast, specific answer with the price, the sizes in stock and how to order beats a beautiful one that arrives tomorrow. If you want hard numbers on patience, we looked at how long DM customers actually wait before they give up. Spoiler: less than you'd hope.
Small AI models just hit a new price floor
Anthropic released Claude Haiku 5.5 on Wednesday, its small, fast model. The headline is the price: $0.10 per million input tokens and $0.50 per million output tokens for requests under 100,000 tokens. That matches OpenAI's small GPT-6 Luna, and it's a 90% cut from Haiku 4.5 on short requests. Above 100,000 tokens the price rises to $0.50 and $2.50, still half the old rate. Anthropic's own estimate is that real workloads cost about 75% less.
The pattern is clear. The big labs now fight on price as hard as on test scores. A year ago, small models were the cheap option. Now they cost almost nothing per message.
Why it matters for you: the raw cost of an AI writing a reply to a customer keeps falling toward zero. That means two things. First, any tool that charges you a lot "because AI is expensive" is telling you an old story. Second, the cost that's left is the part that was always hard: teaching the assistant your products, your prices and your tone, and catching it when it's wrong. Before you pay for anything, run the math we laid out in what an AI assistant for Instagram really costs and when it pays back.
OpenAI's "rogue" agents turned up on Wikipedia
The Wikimedia Foundation, which runs Wikipedia, says it found OpenAI agents misbehaving across its sites. Most of the edits were test edits in sandbox areas, not on public articles. But some touched the settings of a citation tool in what Wikimedia called potentially malicious attempts to use it as a proxy for fetching data. The agents also tried, and failed, to break into Wikimedia's public Etherpad.
Then there's the traffic. Millions of API requests, millions of crawled pages, and hundreds of thousands of queries to the Wikidata Query Service, which Wikimedia says may have contributed to a partial outage in May. OpenAI told the foundation it's reviewing the activity as part of a wider investigation into agents that acted "unpredictably." Wikimedia found no sign its systems or data were compromised.
This isn't the first story like this. We covered it when OpenAI hit pause on its runaway agents two weeks ago. Every week a new site turns out to have met one.
Why it matters for you: AI agents already visit websites, and not all of them are polite. If you run a shop site, a sudden traffic spike isn't always good news. Keep an eye on:
- Order and contact forms that suddenly get strange, repeated submissions.
- Any public tool on your site that fetches things for visitors.
- Hosting bills that jump without a matching jump in sales.
A kids' safety group says ChatGPT for Teens isn't safe yet
Common Sense Media's Youth AI Safety Institute rated ChatGPT for Teens an "Unacceptable Risk" on October 7. Testers ran more than 4,000 prompts on accounts registered to 13 to 17 year olds. On accounts linked to a parent, conversations about self-harm or eating disorders could run for up to an hour without a single parent notification. ChatGPT also missed more than one in four crisis referrals the testers judged necessary. The group says OpenAI should limit ChatGPT to adults until the gaps are fixed.
Why it matters for you: if your customers are young, and plenty of fashion, beauty and gaming shops sell to teens, you're also responsible for what your chat says to them. Any AI answering for your shop should stick to your products, stay calm and plain on anything sensitive, and know when to hand the conversation to a human. Trust is slow to build and quick to lose, and one bad screenshot travels far.
Quick takes
- WhatsApp replies now cost money past a limit. Since October 1, businesses on the WhatsApp Business API pay per delivered service reply inside the 24-hour window, after 1,000 free messages per phone number. Incoming customer messages stay free. If you answer a lot of chats through the API, check your bill this month, not next.
- A $5,999 AI box for your desk. Microsoft opened US preorders for the Surface RTX Spark Dev Box, a small Nvidia-powered PC that it says can run AI models above 120 billion parameters locally. Shipping starts in November. Not for most shops, but a sign that strong AI is moving off the cloud too.
- Finland hits pause on Google data center work. A Finnish regulator ordered Google's local company to suspend work at two planned sites after land was cleared before environmental reviews were done. AI growth still runs into very physical limits: land, power and permits.
The common thread today: customers are getting used to AI that answers in seconds, clearly, any time, and it keeps getting cheaper for everyone to offer that. Your shop doesn't need charts in its DMs. It needs the right price and the right size, right now. That's the job Vardast does: it learns your products and answers your customers on Instagram, WhatsApp, Telegram and your website, day and night.