Top AI announcements from 2025: Nano Banana Sora Perplexity’s AI Browser and More

The year 2025 was characterized by a handful of artificial intelligence (AI) launches that dissipated research labs and into mainstream conversation. Google launched its generative AI portfolio into images and language reasoning with Nano Banana, OpenAI became a social-first app for video generation by Sora; Perplexity expanded its browser ambitions with Comet while Meta organized its long-term research under reviving ‘Superintelligence’ division. The most notable of these announcements was the fact that they were not only for what they built, but also how quickly did the technology categories themselves change in the public eye.

Nano Banana

In August 2025, Google introduced Nano Banana as part of the Gemini 2 in a new version of its , Garcinia. The 5 model family, based on AI-powered image generation and editing, is about the same group as 5, which has been working with images. It was released under the name Gemini 2 as a model for . Originally known as Nano Banana, 5 Flash Image became colloquialized due to its viral spread and widespread use in online communities. The model is designed to create images from natural language prompts and also edit uploaded photos using the same interface, according to Google.

The Gemini app on mobile and desktop is based directly inside Google’s Nano Banana, which allows users to create visuals, change backgrounds (adjusting clothing or objects), add multilingual text inside images. The image model, Google says, uses the watermarking system SynthID (a digital signature) that embeds a “digital” into AI-generated images for provenance tracking.

Sora App

In 2025, OpenAI launched the Sora mobile app based on its diffusion-based text-to-video model and became a consumer-facing creative tool for consumers. Users can input text prompts or reference images from sora, which generates short videos with motion and sound. This models creates clips as high as 60 seconds long, simulated movement according to the prompt context.

In the Sora app rollout, a multi-year, $1 billion deal with The Walt Disney Company was signed that allows soran users to add selected licensed Disney characters from franchises like Star Wars and Pixar in generated videos. The partnership was announced by OpenAI and Disney in December 2025 with the release of the collaboration.

This app was released in iOS, and will soon be available on Android, according to OpenAI, which allows users to share generated videos directly within supported social feeds (which made Sora feel more like a media tool than he did with the developer API demo).

Perplexity’s Comet browser

In 2025, it was announced by Perplexity AI that Comet is its first-of-its-kind AI browser. The browser was released in July on Windows and macOS, followed by an Android version of the browser November 2025. Comet, according to Perplexity, combines traditional browsing with an AI agent that can summarise webpages, answer questions inside the browsing context and read links or page content without switching tabs.

Comet is based on Chromium, meaning it inherits compatibility with Chrome extensions and the Blink rendering engine, Perplexity says. It also includes Perplexity’s conversational AI search assistant, which is branded separately from Google or OpenAI models and has been added to the browser.

Gemini 3 AI model

Gemini 3 was launched by Google in late 2025 as the company’s next-generation multimodal language model family, with variants like Gemin 3 Pro, Ge mini 3 Flash and Gemine 3 DeepThink. Gemini 3 Pro is optimised for a variety of complex reasoning tasks across language and images, according to Google; the lower-latency variant known as Ge mini 3 Flash (which has been developed in order to respond faster) are also available.

Gemini 3 models are available via the Google Cloud’s Ge mini API, and power parts of the Ge Mini chatbot experience on mobile devices.

Several months later, Gemini 3 Flash was released to gain attention as one of the low-latency features in Google’s AI Mode rollout, where it appeared in the browser interface for tasks like summaries and AI-assisted search.

Meta’s Superintelligence Lab

In 2002 Meta organized its long-term AI research efforts under Meta Superintelligence Labs (MSL) in 2025. division consolidated teams that had previously been distributed across Meta AI research groups. Despite the announcement by Meta, MSL is focused on advanced AI research, infrastructure and large-scale model development; it was led by Alexandr Wang, who founded Scale AI in the past.

But the lab isn’t a product launch division; Meta said, “the lab has been devoted to research-first unit that was focused on long-term work on AI systems which could eventually surpass human-level capabilities across domains.”