ChatGPT turns 3 – after coming so far, here's what's next (www.techradar.com)

Three years after its November 2022 debut as a bare-bones “research preview,” ChatGPT has evolved from a text box novelty into a multimodal utility powering hundreds of millions of users and a suite of commercial products. The underlying models advanced from GPT-3.5 to GPT-4, GPT-4o (multimodal), GPT-4.1 (million‑token context), and most recently GPT-5.1, while lighter variants like GPT-4o mini serve as cheaper workhorses. Capabilities now include integrated voice, images, audio and video handling, long-context file tools, agents that act on behalf of users, and native desktop/mobile apps. Public stats cited by OpenAI-linked research point to roughly 700 million users and ~18 billion messages per week. Architecturally, voice interactions run on GPT-4o family models, while the deepest text workflows use GPT-5.1 — a pragmatic multi-model stack that balances latency, cost, and capability. For the AI/ML community, the story is less about a sudden AGI leap and more about rapid productization, scaling tradeoffs, and new engineering patterns. The last three years highlight how multimodality, huge context windows, tool ecosystems, and model specialization drive real-world utility — and costs, monetization (Plus, Team, Enterprise tiers) and governance tensions. Key technical implications: persistent hallucinations, safety/cost/latency trade-offs, uneven progress across domains, and likely future paths toward specialized, integrated assistants rather than a single general intelligence. Competition, regulatory scrutiny, and infrastructure expense will shape which advances reach users and how risks around privacy, bias, and control are managed.