01

ChatGPT starts turning answers into interactive diagrams, comparisons and tools

ChatGPT is getting a more interactive way to answer everyday questions. OpenAI’s new GPT-6 release can combine text with diagrams, charts, buttons and small tools, such as a bill splitter or calculator, directly in a conversation. It can also start answering while it continues thinking or searching, adding findings as they arrive. The global rollout began October 7 for Plus, Pro, Business and Enterprise, with Free and Go starting October 8; workplace settings can affect access. This applies to the Chat tab. The models used in Work and Codex are unchanged, and the Pro reasoning option does not support Intelligent UI.

OpenAI · Oct. 7

02

Anthropic releases Haiku 5.5 and lowers selected API costs

Anthropic released Haiku 5.5 for fast, repetitive work such as summaries, classification and small delegated tasks. For prompts up to 100,000 tokens, API pricing starts at $0.10 per million input tokens and $0.50 per million output tokens; longer prompts cost more. Anthropic also halved Sonnet 5.5’s price for reading cached material. This week, Max subscribers will begin receiving $100 or $200 in monthly API credits, depending on their plan, while Team accounts can receive up to $500 pooled. For businesses building agents, the opportunity is to test cheaper models on bounded tasks while keeping stronger models for more demanding work.

Anthropic · Oct. 7

03

Common Sense Media reports gaps in ChatGPT’s teen alerts and learning controls

Common Sense Media’s Youth AI Safety Institute says parents should not rely on ChatGPT’s teen controls as a complete safety net. After more than 4,000 prompts before and after the August teen-mode launch, it reported failures in parental crisis alerts, age estimation and learning restrictions, while acknowledging that some protections worked. Its overall rating was unacceptable risk, and it called for limiting the product to adults until the gaps are fixed. These are the institute’s findings from specific tests, not a measured failure rate across all users. The assessment examined the August teen experience; it should not be portrayed as a test of October 7’s GPT-6 rollout.

Common Sense Media Youth AI Safety Institute · Oct. 7

04

Unity announces a browser-based game maker connected to its artist asset store

Unity and Google announced a browser-based route into game creation for people without coding or Unity experience. Unity Spark combines prompts with the engine’s game-making tools and connects to the Unity Asset Store, giving creators access to artwork made by its community. The product is in closed beta with a waitlist; the companies describe a broader introduction later this year. Google’s experimental Playground is separately available now to adults. For designers and small creative teams, Spark could make interactive prototypes easier to explore. Unity’s own announcement stresses that an initial generated game still needs creative decisions, iteration and polish.

Unity · Oct. 7

05

Consumers are willing to try AI before they trust it, CapTech finds

Consumers can be skeptical of AI and still give it a try. In CapTech’s survey of 618 U.S. adults, 68.7% said they would use AI that helps them find things, while 35.6% said they would trust it. For systems acting on someone’s behalf, those figures were 47.5% and 27.9%. Respondents also wanted practical controls: 55% wanted to correct errors, and 48.5% wanted access to human review. The survey was conducted in July and August. For a business introducing AI, the findings suggest tracking repeat use and successful outcomes, while making it easy for customers to inspect and correct the system.

CapTech · Oct. 6

Anthropic’s October 7 Haiku 5.5 launch lowers API prices for repetitive work such as summaries and classification. That makes a neglected business question worth revisiting: which useful service have you skipped because preparing it for each small customer took too long?

Consider an independent IT consultant whose five-person clients struggle to follow changes to their booking, invoicing and collaboration software. A weekly, one-page brief could explain which announced changes affect their actual subscriptions, what needs checking and what can wait. The opportunity is a modest recurring service for customers who rarely receive individual research.

Test demand with GPT-6.1 Sol in ChatGPT Work, where available on your paid plan. Work supplies connected-file access, web research and finished documents. Haiku 5.5 becomes a lower-cost candidate if you later build API-based delivery. Its API price cut does not reduce your ChatGPT subscription.

Choose five existing clients with different software needs. Let Work retrieve their approved, anonymized app inventories through your company’s connected files, then check official vendor announcements. A feature for an enterprise plan belongs in a different category from a deadline affecting the plan a client actually uses. Unknown versions or subscriptions stay flagged.

Check every recommendation against its source. Record your review time, including corrections. At a hypothetical $49 monthly price and $60 hourly value for your time, ten minutes of weekly review costs $40 over four weeks. That leaves $9 before tools and overhead. Cheap model calls cannot rescue an expensive review process.

After approving the samples and outreach, offer a four-week paid pilot. Set your test beforehand: at least two of five clients agree to pay, each receives one useful, verified action, and average review takes five minutes or less per client each week. Pause if a brief misstates a deadline or you cannot verify plan eligibility.

For an owner, the decision is whether this deserves a recurring service line. Keep delivery human-approved; schedule private draft preparation only after the first batch passes review.

Today, pick the five clients and price your review time before asking the agent to build the first batch.

AI TOOL SPOTLIGHT · Ownfeed

Ownfeed finds public conversations your product could help with

Ownfeed reads your product’s website, then uses AI to find relevant public posts on Reddit, X, Bluesky and Hacker News. It puts those conversations in a daily feed for you to review and answer yourself. That could help a solo founder test whether people are describing the problem a product solves. Check each original post and the community’s promotion rules before replying. The $1 one-day trial automatically becomes a $39-per-month subscription unless canceled; taxes may apply.

Ownfeed · Oct. 7

People, customers and work

Business tools and services

Models, hardware and operating costs

Research, expertise and safeguards