Time Reveals the Best Inventions of 2025: OpenAI Gets Zero Selections, Domestic Products Dominate the List

Today, Time magazine announced its list of the "Best Inventions of 2025," featuring 300 of the most notable innovations of the year.

Among the inventions on the list are the popular domestic AI model DeepSeek R1, robots that can do the dishes and household chores for us, real-time translation earphones, and small AI supercomputers that bring data centers to the desktop...

These inventions, covering nearly 40 categories, are like a trailer for future life, hinting at how our daily lives will be reshaped.

AI is no longer just a tool for the few; it's becoming ubiquitous;

Robots are no longer confined to laboratories; they're appearing in our kitchens for real;

In fields such as healthcare, agriculture, and entertainment, seemingly minor innovations are quietly changing our daily lives.

Without further ado, let's take a look at the inventions in the world of 2025 that are capturing the world's attention.

It's All About AI

When it comes to innovation, AI is definitely the buzzword. Smartphones have evolved into AI phones, and all the apps on them now come with AI capabilities. In our daily study, work, and life, we're increasingly encountering AI-powered smart healthcare, smart cities, and various AI assistants.

And if we're talking about products that make AI accessible to everyone, DeepSeek R1 is a name that can't be missed.

DeepSeek R1: A Low-Cost Large Language Model

DeepSeek R1 is an open-source inference model released earlier this year on the day of Trump's inauguration. Its performance was almost on par with OpenAI's o1 model, a global AI giant at the time. This sudden emergence of DeepSeek completely disrupted the global AI competition landscape.

Most of the news about DeepSeek at that time was about where it was being integrated. Image source: zaobao.com

The biggest advantage of DeepSeek R1 is that it significantly reduces costs while maintaining stable inference performance. DeepSeek claims to have spent only $6 million on training, far less than the computing resources used by its American competitors like OpenAI and Meta. Moreover, DeepSeek is still available for free use.

Expectations for DeepSeek have been a key theme throughout the year. Compared with other large model providers, DeepSeek's updates can only be described as "slow and steady wins the race." After the initial R1 update at the beginning of the year, there was an update in May, and then there were no more updates to the inference model, except for some partial updates to V3.

While domestic models have made it onto the list, foreign models are also represented. However, this year, OpenAI had zero inclusions; none of its work made it into the Best Inventions of the Year.

Claude Sonnet 4: A Faster and More Accessible Model

Instead, the Claude Sonnet 4 model, released by Anthropic in May this year, made the cut. In Time magazine's citation, it reads:

After its release, Sonnet 4 quickly became a favorite among programmers, capturing more than twice the market share of OpenAI among enterprise developers.

Additionally, Anthropic extended Sonnet's context window, enabling it to handle longer prompts, such as code, up to 75,000 lines, more than twice the scale of its competitors (like OpenAI's GPT-5).

In addition to general models, the popular world models also made it onto the list of the Best Inventions of 2025.

Google DeepMind Genie 3: Creating Virtual Worlds

Unlike ordinary video generation models, world models typically output an interactive virtual environment. We've previously covered the details of the Genie 3 model.

Currently, it's not yet available to the general public; access is limited to a select group of scholars and creators. Potential uses for Genie 3 include education, allowing students to explore a virtual ancient Roman world, or serving as a virtual training environment for autonomous driving and other AI tasks.

Beyond the models, the numerous invention products based on them and the computers used to train them continue to capture our attention.

Nvidia DGX Spark: A Desktop AI Supercomputer

This personal supercomputer, about the size of a Mac Mini, runs on the Ubuntu Linux system. It packs AI computing power into a small device with 128GB of memory and the ability to fine-tune models with up to 200 billion parameters, moving it from expensive cloud service clusters to the desktop.

Dion Harris, the senior director of high-performance computing and AI factory solutions at Nvidia, once said, "Providing petaflops (a quadrillion floating-point operations per second; for comparison, the NVIDIA GeForce RTX 4090 can do 82 trillion per second) of AI performance on a desktop device is groundbreaking because we truly believe it will unlock the democratization of AI."

Currently, this computer is priced at $3,999 and can be pre-ordered through specific sellers. You can also buy two and connect them to handle larger models.

With the best models in hand and advanced devices for fine-tuning, developers are coming up with some incredibly creative and efficient inventions.

Cursor: The Best Practice for Vibe Coding

Without Anthropic's excellent code model, Cursor might not have been born.

The Cursor application interface

Cursor is at the forefront of vibe coding. By combining its in-house developed AI model with models from companies like OpenAI and Anthropic, Cursor can automate a large portion of software development work.

According to Anysphere, the company behind Cursor, more than 50,000 enterprises, including over half of the Fortune 500 companies, are using Cursor, generating over 100 million lines of code every day.

When it comes to vibe coding, Cursor has become a representative product.

Squarespace: An AI-Powered Website Builder

There are many tools for vibe coding, especially those that claim to "build a website with just one sentence."

When I searched for Squarespace on Google, the top three sponsored results were its competitor websites, including bolt.new, mgx.dev, and herclues.app, all of which are AI website builders.

Time magazine mentioned that Squarespace's AI-guided builder, Blueprint AI, helps users generate unique web designs instantly by asking a few questions, rather than starting from scratch. For a more personalized design, it also includes some images curated by the company's own designers.

Paul Gubbay, the chief product officer at Squarespace, said:

Great design is inherently human-centered

We want to ensure that users feel that AI is enhancing them, not replacing them.

Adobe Podcast Enhance Speech: A Tool for Podcasters

Experience it at: https://podcast.adobe.com/enhance

If you've ever thought about recording your own podcast, Adobe's Enhance Speech tool can help you remove noise, echo, and distortion in real-time when recording audio in noisy public places or rooms with poor acoustics.

Enhance Speech has already enhanced over 100 million files.

In a world full of AI voices on the internet, choosing to use AI only to enhance your own voice is quite rare.

Superfluent: Learn a Language Through Conversation

Screenshot of the Superfluent app. You can download it from the app store to experience it.

Superfluent is built on a combination of OpenAI and Google models. It uses conversational AI to assess users' language proficiency and create suitable conversation scenarios, such as ordering coffee at a café or inviting friends to a party.

Superfluent isn't the only product on the list that applies AI to education. There are also well-established plagiarism checkers like Turnity, which, with the help of AI, assist teachers in better controlling how students use AI and understanding their usage.

Another notable invention is the domestic Squirrel AI, a platform that seems to have branches all over the country.

Time magazine describes it as follows: Unlike other AI learning tools that partner with schools, Squirrel AI operates its own learning centers, using adaptive AI models to customize course plans and tests for each student. It has tutored over 24 million students in China through AI and is now expanding into the United States.

Healthcare often goes hand in hand with education, and healthcare-related inventions make up the largest category in this year's Best Inventions list. There are over 46 inventions under the "Healthcare & Wellness" and "Health & Fitness" categories alone.

In the field of AI healthcare research, aside from stories like "ChatGPT saved my life," "AI replacing radiologists," and "AI treating cancer," there are also some practical healthcare popular science examples. Here are some examples:

Outcomes4Me: Demystifying Cancer Care

This is a free AI-based app that can translate medical records and genomic data into easy-to-understand care pathways, clinical trial matching, and symptom tracking tools.

Screenshot of the Outcomes4Me app. You can download it from the app store to experience it.

The creators of this app mentioned that Outcomes4Me allows ordinary people to truly access and understand cancer treatment options.

Currently, over 400,000 patients are using the platform, and it has been rated as the highest-quality cancer app by the peer-reviewed journal Current Oncology.

Phia: An AI Shopping Assistant

When most agent products are promoting their ability to book restaurants and do shopping, and we can even shop on ChatGPT, is it innovative to launch an AI shopping assistant for price comparison?

Screenshot of the Phia app. You can download it from the app store to experience it.

Phia was co-founded by Phoebe Gates (Bill Gates' daughter) from Stanford University and her classmate Sophia Kianni. Its main function is to help consumers decide when to buy products by comparing prices across different platforms