A billion users, a price scissors, and a token meter on one table
The Gemini app's 1 billion monthly users, US-China price competition, GitHub's per-model token breakdown, and one investor's 300-500 million daily tokens together summarize how scale and cost are being measured.
Artificial Intelligence··Morning
A scale milestone and measurement fog
Google chief executive Sundar Pichai announced on X that the Gemini app has passed 1 billion monthly active users, making it Google product number 14 to reach that mark. TechCrunch notes that the figure covers the standalone app only and that Google separately reports 1 billion monthly users for AI Mode in Search. Weeks earlier, on its second-quarter earnings call, the company had reported 950 million monthly users for the app. Google also shared usage figures: 63 percent of Gemini users talk through voice, the app generates 150 million images a day, and it has 100 million active users on iOS. TechCrunch sets the announcement against OpenAI's ChatGPT, which reached 1 billion monthly active users in June 2026, and notes the timing shortly before Google's Made by Google event. Monthly active user counts are company-reported and methods are not published, so it is not certain the two figures are measured the same way. A billion-user threshold draws attention, yet comparison power stays weak until methods are disclosed.[1]
The price scissors cuts both ways
The second meter is price. An AFP report published by The Star describes users shopping between American and Chinese models as cost-efficient Chinese systems press Silicon Valley on price. OpenAI has cut fees by 80 percent for its lightweight model Luna, and Anthropic has released a model approaching its most powerful system at half the price. DeepSeek, whose V4 Flash tops the usage leaderboard on OpenRouter, said it plans a significant increase in prices for programmers. In Zhongguancun, Beijing's technology district, AGI Bar offers free DeepSeek access from two Nvidia workstations self-hosting V4 Flash, so the owner pays the startup nothing. Owner Song De told AFP that providing free tokens is like providing free Wi-Fi. Analyst Jack Gold of J.Gold Associates said a lot can be done with older, smaller or lesser models, especially for agent work. AFP's picture is less a single model winning than buyers moving down from the most expensive frontier systems. Price is a filter that decides which work moves to which model tier.[2]
Token breakdown and a daily consumption claim
The third and fourth meters sit inside the spend breakdown. GitHub says its AI usage report now shows a per-model breakdown of tokens behind AI credits, listing input, output, cache read and cache write tokens beside each model's credits. Before the change, the report showed credits without token detail. The breakdown is available to administrators on Copilot Business and Copilot Enterprise, and to everyone on Copilot for individuals. GitHub says a charge was hard to explain or reduce without seeing which tokens produced it; the new view lets an administrator trace how those tokens add up to a model's cost. The breakdown is obtained by downloading the AI usage report from billing settings, not shown live in the interface. GitHub published the change on the same day it announced MAI-Code-1.1-Flash in Copilot. Rest of World interviews an investor in India's AI ecosystem who says they consume 300 to 500 million tokens a day, spending a few hundred to a few thousand dollars daily, much of it subsidized. OpenAI's Codex and Anthropic's Claude account for roughly 90-95 percent of that usage; Rest of World presents the figures as interview claims, not measured data. Together the four sources read usage like a water meter: tap opening, tariff, and drop-by-drop ledger.[3], [4]
