Covers October 5 to 9, 2026. Checked October 9.
Monday’s roundup was all about AI coding tools. Since then, the big chat apps all shipped something. Here’s what happened, in plain English, and whether you need to do anything.
ChatGPT: GPT-6 arrives, and answers you can click
On October 7, OpenAI started rolling out GPT-6 in ChatGPT, everywhere in the world.
- Paid plans get GPT-6 Sol first.
- Free and Go users get a lighter version, GPT-6 Luna.
The more interesting part is something OpenAI calls Intelligent UI. “UI” means user interface: the buttons and screens you use. Until now, ChatGPT mostly answered with text. Now an answer can include things you can tap and play with: buttons, a little calculator built just for your question, or a chart you can edit.
Example: ask “how much will this car loan cost me?” and instead of a paragraph of math, you might get a small calculator where you slide the interest rate up and down.
Why you’d care: this is a quiet shift from “AI that talks” to “AI that builds a tiny app for your question.” Claude has been doing something similar with artifacts, and now it’s becoming normal everywhere.
If you code with Codex: nothing changes under the hood. OpenAI says the models behind Codex stay the same. Codex did get faster steering in the ChatGPT desktop app (October 8), so when you tell it “no, do it this way” mid-task, it reacts sooner. OpenAI also kicked off a “28 days” run of small daily Codex improvements, starting with a speed boost.
(OpenAI: GPT-6 and Intelligent UI, TechCrunch, 9to5Mac, Codex changelog)
More from OpenAI this week
- Invisible watermarks in Europe (Oct 5). Text that ChatGPT and Codex write for people in the EU will carry a hidden marker that says “AI wrote this,” to meet the EU’s AI Act. You won’t see it, and it’s not happening outside the EU for now. (TechCrunch)
- Propaganda accounts banned (Oct 8). OpenAI caught Russian and Iranian groups using ChatGPT to write propaganda and shut them down. (NPR)
- Teens. Teen accounts got a College Planner, flashcards and quizzes. (OpenAI)
- Ads are coming to image generation. Press reports say OpenAI will test ads inside ChatGPT’s image tool later this month.
Claude: a much cheaper small model, plus dashboards and animations
Claude Haiku 5.5 (Oct 7)
Haiku is Claude’s small, fast, cheap model, the one apps use when they need lots of quick answers. The new Haiku 5.5 is dramatically cheaper than the last one:
| Haiku 4.5 | Haiku 5.5 | |
|---|---|---|
| Reading your prompt (per million tokens) | $1.00 | $0.10 |
| Writing the answer (per million tokens) | $5.00 | $0.50 |
Those prices are for normal-sized requests. Very long ones (over about 75,000 words) cost more. Anthropic says on average it costs about 75% less to run.
It’s also the first Haiku with an effort setting, so a developer can tell it “think hard on this one” or “just answer quickly.”
Anthropic also cut the price of cached reads on Sonnet 5.5 in half. (Caching means Claude remembers the parts of your prompt that repeat, so you don’t pay full price for them every time. We covered it in our prompt caching post.)
Why you’d care: if you pay per token (see our “subscription or pay per token” post), cheap-but-good models are where the savings are. If you only use the Claude app on a subscription, you won’t notice a price change.
(Let’s Data Science, iThinkDiff)
Dashboards and Motion (Oct 8)
Claude can now build dashboards that stay up to date from your company’s data, and Motion turns a report or chart into a short animated video. Both are in beta. At the same time, Claude’s Docs, Slides and Design tools came out of beta and are now on every plan, including Free. We wrote a full post on this one.
(Anthropic: Dashboards and Motion)
Free Claude for startups
Anthropic’s startup program now gives qualifying startups a free year of Claude Team (up to five seats) and $1,000 in API credits. If you’re building something small, it’s worth a look. (TechCrunch)
Google: a Gemini agent that does office work
On October 8, Google Cloud launched a Gemini agent for work. You give it a goal (“put together a summary of last quarter’s customer complaints and draft a reply plan”), and it plans the steps and hands back finished work.
- It works inside Gmail, Docs, Sheets and Calendar, and also through Microsoft 365 and Slack.
- It picks a model for each job. Right now that means Google’s Gemini models and Anthropic’s Claude models, with more to come.
- Versions for finance and legal work are in preview, with government, healthcare and retail coming later.
Why you’d care: the word of the year is agent: AI that doesn’t just answer, it goes and does the task. Google, OpenAI, Anthropic and Microsoft are all racing to be the one that does your office work. Also notice that Google’s agent uses a competitor’s model. The companies compete, but the models get mixed and matched behind the scenes.