Two Assistants, Two Billion Users — and the Model Retirement Wave Begins

This week gave us a genuine milestone: for the first time, two consumer AI assistants have each crossed a billion users, and they got there within weeks of each other.

ChatGPT and Gemini each cross 1 billion active users

ChatGPT hits 1 billion users. OpenAI confirmed that ChatGPT crossed 1 billion active users on July 31 — the fastest consumer software platform in history to reach that mark, doing in under four years what took Facebook more than eight. The company used the moment to cut GPT-5.6 "Luna" API prices by 80%, from $1.00 to $0.20 per million input tokens, while also expanding unlimited free ChatGPT access. The timing isn't a coincidence — OpenAI is widely expected to lean on user-growth numbers ahead of its anticipated IPO.

Gemini isn't far behind. Google's Gemini crossed 1 billion monthly active users on August 12, becoming Google's fastest-growing product ever and its 14th product to hit that milestone. CEO Sundar Pichai said on Alphabet's earnings call that 45% of Google Search queries now surface an AI Overviews component powered by Gemini 3.5 Flash, up from 25% six months ago.

The model retirement wave begins

While user counts climb, the backend of the industry is quietly clearing out old inventory. Providers are retiring older model checkpoints at an unprecedented pace — driven by the reality that keeping dozens of legacy model variants warm on high-demand GPU clusters is simply too expensive.

OpenAI announced deprecation dates for several early 2025 models, including GPT-4o-mini and GPT-4.5-preview, nudging developers toward GPT-5.6 Sol and GPT-5-mini. Anthropic gave developers 60 days' notice on legacy Claude 3.5 Haiku endpoints. For developers building long-term products, the message is clear: model lifecycles are shrinking to 12–18 months, and building robust abstraction layers is no longer optional.

Anthropic locks in Claude Sonnet pricing

Amid the shifts, Anthropic made a developer-friendly pledge this week, guaranteeing that API pricing for Claude 3.7 Sonnet will remain locked through at least Q4 2026. The move aims to reassure enterprise customers and startup builders who have voiced concern over sudden model pricing changes across cloud providers.

EU watermarking goes live, Grok 4.6 ships, and FrontierMath benchmark

On the regulatory front, the EU AI Act's mandatory watermarking and synthetic content labeling rules officially went live on August 12, requiring all major AI platform output in Europe to feature machine-readable watermarking.

Meanwhile, xAI quietly rolled out Grok 4.6 to desktop and enterprise tiers, boasting real-time web grounding and enhanced multi-file code editing. On benchmarks, the new FrontierMath evaluation suite reported frontier models achieving a record 92% accuracy on complex advanced mathematical proof generation.

The bigger picture

As two consumer AI assistants achieve scale equal to the internet's largest platforms, the industry is entering a new phase. Efficiency, model lifecycle management, pricing stability, and regulatory compliance are taking priority alongside raw reasoning capability.