OpenAI published its global ChatGPT usage report with data from the second quarter of 2026, and the regional finding is more specific than one might expect: the three countries that climbed the most positions in the ranking of messages per capita are Peru, Uruguay and Costa Rica. They are not Latin America’s large markets, but three small, middle-income economies. The report covers 126 countries and notes that parts of Latin America, Oceania and Africa grew faster than the rest of the world.
Two figures help show what people are doing with the tool. In work conversations, people are more than twice as likely to ask for a task to be carried out (writing, editing, programming, analyzing data) as they are outside work. And the fastest-growing use case is generating and analyzing multimedia: 7.8% of messages globally, and more than 10% in Brazil and Colombia. The methodology is stated: about 300,000 messages a month between July 2024 and June 2026, classified automatically on de-identified data and normalized by population using World Bank estimates. The limitation is also stated: the figures are published by the company being measured, without external audit and without raw data. They work as a signal, not as official statistics. The same day, OpenAI removed the cap on text chats in its free plan, a change that mechanically reinforces the trend the report describes.
The timing matters because it comes two days after the World Bank’s World Development Report 2026, devoted to artificial intelligence and developing economies. That report measures 4.5% of jobs at risk from generative AI in low- and middle-income countries, compared with 14.2% in high-income ones, and 16.2% of jobs with potential for improvement, compared with 18.7%. In other words: less to lose and almost as much to gain, but all of it conditional on closing gaps in electricity, connectivity, skills and institutions. Its director, Gaurav Nayyar, summed up the point in the official press release: “the window to get this right is narrow.” A caveat is in order: the press release does not break down figures for Latin America and the Caribbean, so before citing it as regional evidence you have to go to the body of the report. High adoption, pending infrastructure. Usage is measured by the company that sells the product; the gap is measured by a multilateral body. Neither measurement is made by the region.
Also today
- SAP freezes travel and most hiring because of the rising cost of AI — An internal memo allows exceptions only for artificial intelligence initiatives, while the company rolls out an AI tool across the organization. AI has not yet replaced anyone and has already eaten the budget that funded hiring; SAP runs service centers in Brazil and Mexico.
- Decade raises $85 million, the largest seed round in Latin American history — A Brazilian wealth advisory firm built on AI, founded by former Nubank CTO Vitor Olivier together with Felipe Meneses, with Greenoaks, Benchmark and Diffusion. It came out of stealth on August 4, still without a launched product.
- Mirendil signs a multiyear deal worth more than $100 million with Google Cloud — A self-improving AI lab founded by former Anthropic staff, with access to TPUs, Nvidia GPUs and managed clusters, after raising about $200 million at a $1 billion valuation in June.
- Naïve raises $28.5 million so AI agents can incorporate and run companies on their own — A single programming interface for payments, email, phone, cloud, storage and incorporation. The first jurisdiction to be automated is the U.S. limited liability company.
- OpenAI’s smart speaker designed by Jony Ive would cost between $300 and $400 — A metal donut with moving parts, at the price of a mid-range phone, on the same day that talking to that model became free and unlimited.
In the region
The most concrete institutional move came from Brazil. Portaria TSE nº 463/2026 sets August 16 as the deadline for platforms with more than 5 million users in the country (individually or as an economic group) to submit electoral compliance plans, and for the first time in the region the requirement explicitly includes that generative AI providers describe the technical safeguards that prevent their systems from recommending candidates or producing non-consensual sexual material of candidates, as well as the mechanisms for identifying and labeling synthetic content. Laís Martins’s analysis in Tech Policy Press points to the issue that decides everything: the plans have a public layer and a restricted one, and it is not clear that an electoral court has the technical capacity to check what is declared against what the system actually does. The contrast with Colombia, which reached its June runoff without an equivalent institution and with Law 2502 lacking active sanctions, remains the best argument in favor of the Brazilian mechanism; August 16 will show whether paper turns into enforcement. No multilateral body with a regional mandate published anything in this edition’s window. Meanwhile, two decisions made abroad affect the region without consulting it: OpenAI partnered with the American Psychological Association to define how ChatGPT should behave with teenagers ages 13 to 17 (the product is the same one Latin American teenagers use, and no professional society from the region is at that table), and Suno announced watermarking and fingerprinting for its generated music, without publishing the technical specification that a radio station, a court or an electoral authority would need in order to verify anything.
Launches
- ChatGPT with unlimited text chats on the free plan and GPT-5.6 Luna — The Free and Go plans move to uncapped text conversations, with GPT-5.6 Luna replacing GPT-5.5 Instant and a “Think” button for complex questions; Plus and Pro get a GPT-5.6 Sol with a slider that regulates how much the model reasons per response. OpenAI reports 62% fewer factual errors with Luna and 68% fewer with Sol compared with the previous model. Caps remain on files, images, voice and image generation. Sol is already available; the unlimited chats and the “Think” button arrive within a week. It is the change with the most direct effect in the region, because it removes the barrier that weighs most where paying in dollars is not an option.
- Claude Fable 5 with recalibrated biology classifiers — Anthropic cut refusals on biology queries by about 85%, with overall reductions of 67% on Claude.ai, 55% on Cowork, 17% on Claude Code and 7% on its developer platform; it still blocks virology, toxicology and molecular design. It rolls out without users having to do anything. It matters to anyone making clinical, educational or public health queries who had been running into unexplained refusals.
- Ask Maps with agentic actions and Personal Intelligence — Google Maps adds food ordering with Square, Toast and Uber Eats, hotel booking with price comparison and ticket search, plus conversational memory, a live public transit widget and integration with Gmail and Calendar turned off by default. The transactional features are U.S.-only; the region gets the memory and the transit widget. It is free within Maps, the Google app with the deepest real penetration in Latin America.
Threads we’re following
Yesterday we reported that the new U.S. federal safety review framework leaves out open-weights models, precisely the ones the region deploys for price and for data sovereignty. Today’s World Bank report supplies the other half of that equation: the potential for job improvement in developing economies is almost as high as in rich ones, but it depends on electricity, connectivity, skills and institutions. It is the same thread seen from two angles. The technical capability arrives (open, cheap, increasingly free), and what does not arrive with it is someone to evaluate it, someone to measure its effect and someone to answer for it when it fails.
If tomorrow a ministry needs to justify an investment, a law or an artificial intelligence training program, it will cite OpenAI’s ranking, because it is the only thing there is. What would it take for the region to have its own adoption figure, and who would have the legitimacy to produce it?
About this entry. It is generated automatically from public sources, 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 or making decisions based on it.
Doble Click is written with Anthropic models.