From Nano Banana’s rise to agentic AI misfires — here’s what defined 2025 in AI

Looking back at the highs and lows of AI in 2025

28 Dec 2025, 11:56 · TechRadar

Features By Graham Barlow published 28 December 2025

(Image credit: Getty Images/Alexsl) Share Share by:

Share this article 0 Join the conversation Follow us Add us as a preferred source on Google

As the Senior AI Editor on TechRadar I’ve seen AI use grow throughout 2025, with some really nice apps and features getting released. My particular highlights have been the Purpose app from Marc Manson, and the marvellous Nano Banana image generator from Google.

2025 started with a bang with the release of DeepSeek R1, the Chinese AI model that rivaled the power of ChatGPT, but at a fraction of the cost for developers. With a start like that it looked like 2025 was all set up to be the year that AI ‘arrived’ on the world stage, with even the possibility of Artificial General Intelligence (AGI) being achieved.

AGI was billed for 2025 by many, including Elon Musk, but it simply hasn’t materialized, although the year has still been a strong one for companies like OpenAI, xAI, Anthropic, and Google. That said, there have also been some massive flops along the way – and as for Apple, it feels like yet another year of slipping further behind in the AI race, with a revamped Apple Intelligence now due to arrive some time in 2026. In contrast, Google and Samsung have been pushing AI further in mobile devices than ever before.

OpenAI learns a lesson

ChatGPT has maintained its vice-like grip as the most popular AI chatbot in the world, although it hasn’t been plain sailing for the tech giant. Legal challenges, particularly the copyright infringement claim brought by The New York Times, have continued to dog the company, and in June its servers crashed for a couple of days, giving the world a brief taste of life without the ubiquitous chatbot, and users were not happy.

OpenAI then fumbled the ball with the release of the GPT-5 model, which came across as cold and unemotional compared to the previous GPT-4o. The shift disappointed millions of users who had come to rely on the chatbot as something closer to a trusted companion. It felt like a best friend had undergone a personality transplant overnight, forcing OpenAI to make the legacy 4o model available again.

The company has also lost a little ground to Google’s Gemini in recent months. The arrival of Gemini 3 Pro in November was well received, and on the image front Gemini’s Nano Banana and Nano Banana Pro proved superior to ChatGPT for image generation. OpenAI responded with a new image-generation model in December.

2025 was also supposed to be the year that AI agents — AI assistants capable of working autonomously on tasks for us — entered the workforce, and while we did see some impressive releases, like OpenAI’s Agent Mode and Perplexity’s Comet Browser offered agentic browsing, we still aren’t living in a world where AI agents are doing tasks for us regularly. The problem is that they can still make little mistakes, and until AI can perform each task perfectly we can’t trust it to perform any task completely.

Get daily insight, inspiration and deals in your inbox

Contact me with news and offers from other Future brandsReceive email from us on behalf of our trusted partners or sponsors