Anthropic released a cheaper Claude Haiku 5.5 and widened cyber access for vetted defenders, OpenAI rolled GPT-6 out to all ChatGPT users while a watchdog rated its teen version an unacceptable risk, Google narrowed free Gemini access, Meta detailed child-safety measures and new ad tools, and xAI's Grok Bot began routing some work to rival models.

Anthropic releases Claude Haiku 5.5 at much lower API prices

• Anthropic launched Claude Haiku 5.5, a smaller model it says is aimed at cost-sensitive tasks.
• VentureBeat reports a 90% API price cut, and Channel NewsAsia reports pricing starting at $0.10 per million tokens.
• Reuters says it is the third Claude 5.5 model and arrives ahead of Anthropic's planned IPO.

For you Developers running high-volume, cost-sensitive work on Claude's API have a cheaper model to test.

Anthropic opens stronger cyber capabilities to more vetted security teams

• Anthropic expanded its Cyber Verification Program, giving qualifying security professionals advanced cyber capabilities with fewer blocking classifiers.
• The program now has three access tiers, and SiliconANGLE reports it absorbs Project Glasswing.
• MIXED reports the most permissive Red Team tier blocked none of 50 trials in Anthropic's own test.

For you Security professionals who qualify can get Claude access with fewer cyber-related refusals.

Pentagon stops using Anthropic AI after a court ruling

• MeriTalk reports the Defense Department halted use of Anthropic's AI after a court upheld its supply chain risk designation.
• Forkast links the Pentagon's decision to Anthropic's safety guardrails.

Japanese book wholesaler confirms selling books to Anthropic

• A Japanese book wholesaler has admitted selling books to Anthropic, Kyodo News reports, citing an industry source.
• Nikkei Asia reports Anthropic's mass book buying in Japan has drawn a backlash from publishers.

Free Gemini users will be limited to a single model

• Free Gemini users are being limited to one model and will lose access to Flash and Pro, several outlets report.
• Inshorts reports the change takes effect in India from October 9.

For you If you use Gemini without a subscription, expect fewer model choices.

Google launches Playground for creating and sharing custom games

• Playground is a new experimental platform for creating, playing and sharing custom games.
• The Decoder reports Google is aiming it at casual players who want to make games with Gemini.

For you Gemini users can try making and sharing their own games.

Gemini in Chrome reaches Android in India, with auto browse

• Gemini in Chrome is now live on Android in India.
• Auto browse, which can search, compare and complete web tasks, is rolling out to AI Pro and Ultra subscribers.

For you Android Chrome users in India get Gemini in the browser; auto browse needs an AI Pro or Ultra plan.

Google DeepMind releases EmbeddingGemma 2, an open embedding model

• Google DeepMind released EmbeddingGemma 2, a new version of its open embedding model.
• Google describes it as lightweight and multimodal.

For you Developers who need embeddings can try an open, multimodal option.

Google unveils Gemini Nano Banana 2.1

• Google unveiled Gemini Nano Banana 2.1, which ET Enterprise AI describes as cost-efficient with feature upgrades.
• Digit reports it was announced alongside EmbeddingGemma 2.

Meta details new measures against child sexual exploitation

• Meta published additional measures it has put in place to fight child sexual exploitation on its platforms.
• Yahoo Tech reports they include AI tools to detect ads that secretly lead to child sexual abuse material.
• NewsBytes reports Meta says it acted on 33.2 million pieces of such content globally.

Meta adds AI ad tools and tests an agentic campaign assistant

• Meta introduced new AI-powered tools for advertisers and agencies at Advertising Week.
• BMI reports Meta is testing an ad assistant that can create campaigns and change targeting and budgets.
• TechJuice reports the new tools include in-chat checkout.

For you Advertisers on Meta's apps get new AI tools, with an assistant that can change targeting and budgets in testing.

Meta's Muse comes to the iPad

• Meta's Muse app is now available on iPad.
• Yahoo Tech notes the launch comes about a month after Muse's mobile debut.

For you iPad owners can now use Meta's Muse agent on a larger screen.

Parental supervision for the Meta AI app is available worldwide

• Tech in Asia reports parental supervision for the Meta AI app is now available worldwide.
• This is a second independent report of the feature, which one outlet described yesterday.

For you Parents of teens who use the Meta AI app can turn on supervision.

GPT-6 with a new visual interface rolls out to all ChatGPT users

• OpenAI is rolling out GPT-6 globally in ChatGPT with Intelligent UI, which adds visuals and interactive elements to answers.
• OpenAI says responses are faster; Unite.AI reports it reaches every ChatGPT tier.

For you ChatGPT users, free tier included, will start seeing more visual, interactive answers.

Watchdog rates ChatGPT for Teens an 'unacceptable risk'; OpenAI disputes study

• Common Sense Media rated ChatGPT for Teens an "unacceptable risk" and said it should be paused.
• The report found teen safeguards failed to alert parents during suicide conversations, the Los Angeles Times reports.
• OpenAI contends the study is flawed, and told Reuters teens use ChatGPT for under 15 minutes a day.

For you If your teen uses ChatGPT for Teens, this report questions whether its parental alerts work during crisis conversations.

ChatGPT for Teens adds a college planner and study tools

• OpenAI is bringing College Planner to ChatGPT for Teens to help students manage college applications.
• It also adds flashcards, quizzes and a teen AI council; Fox Business says the planner tracks financial aid too.

For you Teens on ChatGPT for Teens can track college applications and study with flashcards and quizzes.

Radisson launches a hotel booking plugin inside ChatGPT

• Radisson Hotel Group built a ChatGPT plugin with Accenture to help travelers find, compare and book hotels.
• Unite.AI reports OpenAI shared early results from the plugin.

For you ChatGPT users planning trips can search and book Radisson hotels inside the chat.

Grok Bot will route some tasks to Claude, Midjourney and Suno

• Musk said Grok Bot will send tasks to the best model for the job, including Anthropic's Claude Opus 5.5, Midjourney and Suno.
• The Information reports SpaceX will sometimes use rival models to power Grok Bot; Android Headlines says Claude handles heavy reasoning and coding.

For you Grok Bot users may get some answers from non-xAI models, including Claude, for reasoning and coding work.

SpaceX said to seek $40 billion to buy Nvidia chips

• Unconfirmed: the Financial Times and Bloomberg's sources say SpaceX is seeking $40 billion in financing to buy Nvidia chips.
• Livemint reports SpaceX is turning to banks and investors to fund Musk's AI expansion.