1Highlight: Google opens Gemini 4 Carbon, its strongest model, to paying users; price doubles after 30 Nov.
2Businesses can now claim and update their profile inside ChatGPT; agent features are US-only.
3Meta's Muse Ember aims to make routine AI agent work cheaper; most scores are company figures.
Highlight of the day
AI models
US$2 / US$10
per million input / output tokens until 30 Nov, then US$4 / US$20 (Google)
Merged from 3 sources
★ Highlight of the day
Google opens Gemini 4 Carbon, its strongest model, to paying users
On 10 November Google made Gemini 4 Carbon, which it calls its strongest model, available beyond the cybersecurity testers who got it two weeks earlier. Developers can use it through the paid Gemini API Application programming interface: a door that lets one piece of software ask another for data or actions., and Google AI Ultra subscribers can try it in the Gemini app, AI Mode in Search and Google Antigravity Google's software for building apps with AI agents, which can write, run and check code for you., Google's AI coding tool. Most performance claims below are Google's own.
What we learned
Until 30 November the API costs US$2 per million input Token A small chunk of text, roughly part of a word, that AI models read and write; AI usage and pricing are counted in tokens. and US$10 per million output tokens (about RM8 and RM41), the same as Gemini 4 Argon. After that it doubles to US$4 and US$20 (about RM16 and RM81). Google says Carbon uses more tokens than Argon because it explores and checks its work for longer.
Google says Argon is still the right choice for most coding and office work; Carbon is for very hard, multi-step jobs. It says Carbon ranks first on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench at less than half the cost per task of the next best outside model. Google calls these independent tests; we have not seen the full tables.
Google is previewing '/swarm' in Antigravity, which turns one request into a team of AI agent An AI program that can take actions for you, like writing code, filling forms or moving items, not just answering questions. you can watch at work. Carbon takes the planning and checking roles, and faster, cheaper models do the rest.
In a hands-on video, the creator of the Prompt Engineering channel, who says he recently joined Google and was given early access, left Carbon working for about 20 hours on a 3D model of a fort in Peshawar. A fresh AI reviewer scored it about 5 out of 10. His advice: ask for many rounds of work, a progress file and a separate reviewer.
Why it matters: If you're on Google AI Ultra or pay for the API, try Carbon on one long, hard task where checking matters, and compare its cost with Argon before switching. Most people won't need it for everyday writing or chat.
Gemini 4 Carbon is now available · Google (The Keyword)Read ↗
Gemini 4 Carbon is now available for developers · Google (The Keyword)Read ↗
Gemini 4 Carbon: The Most Capable Model from Google · Prompt EngineeringWatch ↗
Was this useful?Tap to vote. Hold, right-click or press Shift+Enter to add a reason or a note first.
More stories this day
Business
1.2bn
weekly ChatGPT users (OpenAI's own figure)
Business
Businesses can now claim and update their profile inside ChatGPT
On 10 November OpenAI launched Businesses in ChatGPT, in beta. Through its Business Hub, a company can claim its listing, correct key details and add product feeds with prices and stock, so ChatGPT has current information when people ask about it. OpenAI says any eligible business located where ChatGPT is available can claim its listing.
Key learnings
Once verified, a business sees aggregated numbers: how often it is mentioned, how many people click through to its website, and common topics in conversations where its site is cited. OpenAI says it does not share people's chats with businesses, and Insights A weekly view for Notion workspace admins of how their team uses Notion AI. To build it, Notion used to send saved AI chats to an outside company to turn each one into a short description. Notion has now switched this feature off. only appear above a minimum number of conversations.
Business AI agent An AI program that can take actions for you, like writing code, filling forms or moving items, not just answering questions., chat assistants a business sets up to answer questions about its own products, are available only to eligible online shops and online-service businesses in the United States for now. Those chats are labelled and kept separate from ChatGPT's own answers.
Businesses can also pay to promote their Business Agent as the landing point for ChatGPT ads; sponsored placements are labelled. OpenAI says 1.2 billion people use ChatGPT each week, its own figure, and plans to expand to more countries later without a date.
Why it matters: If you run a shop or service in Malaysia, ask ChatGPT about your business to see how it describes you, then check in Business Hub whether you can claim the listing and fix prices, hours or products. The agent features are US-only for now.
Was this useful?Tap to vote. Hold, right-click or press Shift+Enter to add a reason or a note first.
AI models
US$0.10 / US$0.50
per million input / output tokens, prompts up to 200k (Meta)
Merged from 3 sources
AI models
Meta's Muse Ember aims to make routine AI agent work cheaper
On 9 November (US time) Meta released Muse Ember, its smallest and cheapest model, built for the small steps AI agent An AI program that can take actions for you, like writing code, filling forms or moving items, not just answering questions. repeat all day: calling tools, small coding jobs, sorting requests and reading long documents. On Meta's Model API Application programming interface: a door that lets one piece of software ask another for data or actions. it costs US$0.10 per million input Token A small chunk of text, roughly part of a word, that AI models read and write; AI usage and pricing are counted in tokens. and US$0.50 per million output tokens (about RM0.40 and RM2).
Key learnings
That is the same headline price as OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5. Independent developer Simon Willison notes the price rises for very long prompts: above 200,000 tokens Ember costs US$0.30 in and US$1.50 out.
Meta says Ember matches Haiku's best score on the Artificial Analysis index at about a third of the cost per task, and beats every Luna setting. Meta's chart cites Artificial Analysis results dated 9 November, though its Ember costs are estimates, and several other test scores are Meta's own.
It is Multimodal Describes an AI model that can understand more than one kind of input, such as text, images, audio and video., reading text, images, audio and video, and takes up to 1 million tokens at once. It has six reasoning levels; in Willison's quick drawing test, higher levels did not reliably give better results, and the 'high' level failed to finish.
Meta says it tested Ember for chemical, biological, cyber and loss-of-control risks and found all below its own high-risk thresholds, and checked whether it follows instructions hidden in web pages. These are Meta's own assessments.
Why it matters: If you or your developers run AI agents on lots of small tasks, a cheaper model for the routine steps can cut bills. Test it on your own tasks first: cheap per token isn't always cheap per job.
AI models
