DeepSeek, ChatGPT Atlas, Edits: Discover the Tools That Defined 2025!

DeepSeek

On January 26, 2025, after creating a massive buzz on social media, DeepSeek climbed to the top of the U.S. App Store charts, becoming the first generative AI service to temporarily surpass ChatGPT in download numbers. This event caused a significant market shakeup, including a 18.6% drop in Nvidia’s stock (Nvidia supplies OpenAI), and also marked a pivotal moment in China’s involvement in the generative AI race. Silicon Valley’s renowned investor Marc Andreessen referred to it on X as the “Sputnik moment of AI.”

Deepseek R1 is AI’s Sputnik moment.

— Marc Andreessen (@pmarca) January 26, 2025

While the “DeepSeek effect” has since faded, and its models now struggle to compete with those from companies like Anthropic or Google, this conversational agent has firmly established itself in the market. As of last August, it was the third most-used generative AI tool on the web and eighth on mobile, according to a study by Andreessen Horowitz. It would not be surprising to see it maintain this status into 2026.

Edits

In a move against ByteDance, Meta launched Edits this year, a direct competitor to CapCut. Initially set for March, the release was postponed to April. Edits offers features similar to its rival, including cutting functions, subtitle generation, green screen overlays, and visual effects. The launch appears to have been successful; according to AppFigures, Edits surpassed 7 million downloads in its first week, briefly becoming the most downloaded app on the U.S. App Store.

However, it’s too soon to claim it competes with CapCut. In France, for instance, ByteDance’s app still ranked eleventh in the most downloaded apps of the year on iOS, a list where Edits does not appear. Yet, the frequency of its updates could potentially get it there. Missing many features at launch, the app now benefits from monthly or bimonthly updates that are gradually strengthening it. In recent weeks, Edits has enhanced its storyboard functionality and integrated new templates. Earlier in the year, it also added an integrated teleprompter.

NotebookLM

NotebookLM, though not launched this year, has seen continued success and improvements that warrant its mention here. It went viral on social media in fall 2024 for its ability to generate incredibly realistic podcasts using multiple documentary sources. The tool developed by Google Labs has seen significant enhancements recently.

Now available in over 80 languages, including French, NotebookLM can generate videos synthesizing multiple sources, interactive quizzes, and study notes. It also includes the Deep Research feature, a capability from Gemini that analyzes and summarizes hundreds of documents swiftly, and will soon introduce the nano-banana image generation model. A busy roadmap indeed.

ChatGPT Atlas

While companies like Perplexity, The Browser Company, and Opera have paved the way with Comet, Dia, and Neon respectively, it is OpenAI that could take the lead in the emerging battle of AI-enhanced browsers with ChatGPT Atlas. It achieves this by offering features similar to its competitors while leveraging the fame of its flagship product.

To persuade steadfast users of Chrome, Firefox, or Safari to switch, OpenAI hasn’t revolutionized the interface. Based on Chromium, its browser features a classic tab bar, a taskbar for accessing favorites, tabs, or settings. Its homepage is limited to a central bar for entering URLs or querying ChatGPT.

The real transformation lies in changing how we browse the web, further integrating its conversational agent into daily use, which California-based company hopes will reshuffle the deck. On Atlas, ChatGPT sidelines Google and becomes the go-to interface for search. It’s also the assistant you can call upon at any moment to summarize a page, rephrase text, or even perform tasks autonomously through Agent mode. But does saving a few minutes justify handing over all your browsing habits to a company like OpenAI? That’s for users to decide.

Reve

Reve might represent the last survivor of an era where the image generation market wasn’t dominated by giants like Google, ByteDance, or OpenAI, but rather by smaller, often independent players. Launched in March 2025, it quickly topped the collaborative ranking of Artificial Analysis, designed to objectively measure the performance of image generation technologies. Reve stood out for its superior prompt adherence, image quality at a reportedly lower cost, and its unique ability to autonomously enhance user queries with its Enhance feature.

With the 1.0 version of its model, Reve managed to outperform — at least according to users of the Text-to-image Arena of Artificial Analysis — 4o Image Generation, which had marked a significant advancement for OpenAI over DALL-E. However, like Midjourney or Recraft, Reve struggles to keep up with the pace set by Google or OpenAI, who have ramped up investments to catch up in this field historically. Its most advanced model currently ranks thirtieth on this closely watched leaderboard.

Sora

In October, OpenAI unveiled Sora 2, the model now powering its video generator, representing a significant leap in modeling and simulating reality and in creating soundscapes. The sequences generated achieve a level of sophistication previously unattainable for competing tools and even for the first version of Sora. “The launch of Sora 2 represents, in our view, a video content advancement equivalent to the arrival of GPT-3.5,” OpenAI had announced at its release.

Even to the trained eye, detecting these AI-generated videos becomes incredibly challenging, with development spread over twenty months, as signs that previously made such videos easy to spot, like deformed hands or awkward physics, gradually disappear. The Caméos feature, allowing users to insert their own or another consenting person’s face into any scene, further blurs the line between reality and artificial.

While not overly concerned about the web eventually being saturated with fabricated content, OpenAI has still implemented some protections to aid in the detection of fake videos and to prevent the proliferation of deepfakes of public figures. But do these really hold up against bypass techniques? And does OpenAI truly care? If it hasn’t already, it’s high time to start worrying.

Figma Make

Launched in beta in May and available since July, Figma Make represents perhaps the most promising product evolution introduced by Figma this year. Powered by Anthropic’s models, this code generator allows users to convert textual descriptions or existing designs into functional prototypes in just a few clicks, enabling them to “quickly explore and iterate on their ideas,” as stated by the publisher during its annual event.

More than just a “simple vibe coding tool,” as Figma’s product manager reminded BDM, Make integrates with other platform modules and leverages component libraries and design systems to maintain visual consistency. Designed for collaboration, it also includes a system of customizable templates, which designers enhance with design directives, brand guidelines, or product specifications, reusable by the entire team. Quite a feat.