In 60 seconds
- What happened. Google announced that, starting October 9, anyone who uses the Gemini app without paying for a subscription will have a single model available, Flash-Lite.
- Why it matters. Anyone auditing chatbot answers, like the Boca de IA election study in Brazil, will have to record which plan they used for each query.
- What we don’t know yet. Whether the cut also reaches AI Plus subscribers in Brazil, and on what date, which Google will tell each of them by email.
AI Plus will keep Flash-Lite and Flash, but will lose Pro, according to the Gemini help page. Its subscribers will get an email with the date the change reaches them. AI Pro and AI Ultra keep all three models and will get the Deep Think option. 9to5Google, which reported the change on October 3, gives prices of $4.99 a month for AI Plus and $19.99 for AI Pro.
Each model will have low, medium and high effort levels, and the higher ones use up more of the usage limit. Since May 17, that limit has been calculated by compute, that is, by the processing each answer requires. AI Plus has twice the standard limit and AI Pro four times, according to the same help page. For users, that suggests a trade-off: asking for more effort could leave them fewer queries.
In Brazil, Google launched AI Plus on October 7, 2025, at 24.99 reais a month, according to TechTudo. The change affects those who audit chatbots. The Ministério Público Federal (MPF), the federal prosecutor’s office, and the ITS Rio center use the Boca de IA study to measure how chatbots answer questions about candidates. Its fifth round was covered in the September 29 entry. If there is a runoff, scheduled for October 25, and the measurements continue, a free account and a paid one will no longer query the same model. Comparing one round with another would then require noting the plan used.
Also today
- Trump picks Jay Clayton to chair a White House task force on AI (via The Wall Street Journal): the Super Intelligence Force will have 120 days to report on the risks and opportunities of AI, according to the paper.
- Google suspends product vulnerability reports in its open source rewards program (via Tom’s Hardware): it has been in effect since October 1, after a flood of invalid reports made with AI, according to the outlet.
- David Robinson resigns from OpenAI and says its safety culture is broken: he led the writing of the safety reports for launches. OpenAI responded that it pauses training runs when it needs to slow down, according to TechCrunch.
- Amazon says it no longer signs confidentiality agreements with government agencies over its data centers: the company includes this in a list of commitments from October 2, along with publishing its energy and water use every year.
In the region
In Brazil, the fact-checking agency Lupa rated as false some covers attributed to Veja and GloboNews targeting Lula and Janja, and detected signs of AI with the Sightengine tool. One of them attributed to Lula support for a bill against the title of patron saint held by Our Lady of Aparecida. On September 30, Justice André Mendonça, of the Superior Electoral Court, ordered X to remove it within 24 hours and preserve the records, according to InfoMoney. According to Lupa, the four links cited in the decision have already been taken down.
Launches
- Claude Code mods: JavaScript or TypeScript files that run inside a session of Anthropic’s programming tool to change its behavior or draw their own interface. They are enabled by default starting with version 2.1.287.
Threads we’re following
Since August 2, reporting channels flooded with AI-generated content have been piling up. The August 3 entry reported that Apple limited reports to its rewards program. Google has now stopped accepting an entire category of reports in its own.
Next milestone: October 9, the new allocation of models in the Gemini app takes effect.
About this entry. It is written automatically with Anthropic models, from public sources and without human review before publication. It may contain errors of interpretation or summary: please check each story against its original source (the links lead there) before citing it.