Google is rolling out its latest version of Gemini and a new artificial intelligence model designed to simulate the physical world, as the search giant races to keep pace in model development while also providing more agentic services to its massive user base.

The company made the announcements at its annual Google I/O developer conference on Tuesday, gaining an audience for new product debuts at a time when the market has been focused on the soaring valuations of OpenAI and Anthropic, which are both gearing up for IPOs as soon as this year.

The centerpiece of Google’s AI strategy is Gemini, its family of models and tools. The company is showcasing Gemini 3.5 Flash, a lighter-weight addition to its suite that offers cutting-edge capabilities at half, or in some cases close to one-third, the price of comparable frontier models, according to CEO Sundar Pichai.

In a news briefing with reporters ahead of Tuesday’s event, Pichai said Gemini 3.5 Flash is “remarkably fast.” The company said 3.5 Flash will now be the default model for the Gemini app and AI mode in search globally.

“You no longer have to trade quality for latency,” Google said in a blog post. The company said that it’s strengthened the cybersecurity defenses for Gemini 3.5 Flash, so it’s “less likely to generate harmful content and mistakenly refuse to answer safe queries.”

Google said Gemini 3.5 Pro, its heavier-weight version, is being used internally, but won’t be ready for wider distribution until next month.

On the agentic AI front, Google announced Gemini Spark, a new general-purpose AI agent in the Gemini app that can reason across information in connected apps. Google said it wants to help users navigate their digital lives by taking “action on your behalf while under your direction.” Gemini Spark is in beta and will be available first to trusted testers and Google AI Ultra subscribers, starting next week.

With more internet users gravitating to chatbots, Google is trying to convince traditional search users that it can be trusted to help them with tasks involving minimal input. Following the company’s skyrocketing capital spending, Wall Street is looking for Google to show it can create deeper integrations across its products, and agents could be a way to do that.

Expectations for AI companies continue to grow, particularly in light of Anthropic’s recently released Mythos model, which was said to be so powerful that it has found thousands of previously unknown vulnerabilities in the world’s software infrastructure.

Google’s AI portfolio now includes Omni, a world model designed to simulate physical environments, predicting what happens next based on a user’s actions. World models are often used in robotics and gaming and have been heavily researched by DeepMind through the years.

Omni will work in Flash, the Gemini App, Google Flow and YouTube shorts, supporting image and audio, the company said, adding in a separate blog post that users can have Omni edit videos and create more realistic imagery.

“Take a video you shot and just ask Omni to change what’s happening,” the post says. The AI can “edit the action, add in new characters or objects.”

WATCH: What to expect from Google I/O

China’s artificial intelligence race has no finish line. DeepSeek, Moonshot AI, Alibaba and even consumer electronics firm Xiaomi have all dropped new models in recent weeks, jostling for position on leaderboards.

From native AI startups to platform giants, companies across the sector face growing pressure to innovate, expand their user base and find paths to generate revenue. At the same time, they must manage steep research and development costs alongside rising expenses for computing power and hardware.

SenseTime, one of China’s early AI companies, has pivoted to stay relevant in the generative AI era. Long known for facial and image recognition, the company now develops multimodal systems that can combine text, audio and visual data.

Founded in Hong Kong in 2014, SenseTime has faced U.S. sanctions over allegations related to surveillance of Muslim minorities in Xinjiang, which it has denied.

Its latest model, SenseNova U1, integrates language and vision processing into a single system, improving speed and efficiency by removing the need to translate different modes.

SenseTime is betting on cost efficiency as a competitive edge. The company has taken cues from DeepSeek’s approach of delivering high-performing models under financial and technological constraints, according to cofounder and chief scientist Lin Dahua.

While ChatGPT Images 2.0, an artificial intelligence tool from OpenAI that generates images from text prompts, produces “exquisite and beautiful” results, SenseNova U1 costs ten times less, Lin said.

“You may not need the top model in many cases when it can handle most tasks,” Lin told CNBC. “There is still a gap between us and the international frontier models like OpenAI’s GPT Image 2 and (Gemini’s) Nano Banana, but our cost is much lower – it’s very efficient.”

With limited overlap between U.S. and Chinese AI markets, the real competition may be closer to home.

ByteDance’s AI video model Seedance posed a competitive concern at first, Lin said. SenseTime has since integrated some of its capabilities into its short-video tool Seko, allowing it to combine Seedance’s background generation with its own audio functions.

Technology is only half the battle, with business models becoming increasingly important. OpenAI’s reported miss on revenue and user targets, according to The Wall Street Journal, signals risks for Chinese and American players alike, Jefferies said in an April 28 note.

Pure-play AI model companies face a tough equation: low customer loyalty, limited differentiation, a crowded field and high training costs, Jefferies said.

Large internet platforms, by contrast, have stronger cash flow, access to user data and established customer bases to sell AI applications to, the bank added.

In China, platform companies, including Alibaba, Tencent and ByteDance, can use their core businesses to subsidize AI development and enhance existing operations, said Vey-Sern Ling, a senior equity advisor at UBP.

“They are obviously in a better position than the standalone ones, which continue to be loss-making,” Ling added, while noting that heavy AI spending has weighed on profits even at larger players like Alibaba and Kuaishou.

To stand apart, SenseTime has combined large AI models, applications and infrastructure to improve service quality while lowering costs per use, Lin said. Many of its products target enterprise clients, who often demand higher-quality services, are willing to pay more and less likely to switch providers.

SenseTime narrowed its net loss by 58.6% last year and reported positive EBITDA in the second half for the first time since listing in 2021 – a trajectory investors will keep watching closely. Lin said the company’s AI costs are “manageable” and largely focused on making models more efficient.

Shares of SenseTime rose 2.5% at Wednesday’s open.

Quantum stocks climbed on Thursday, adding to a massive week-to-date rally fueled by enthusiasm for Nvidia‘s new open-source artificial intelligence models designed to advance the burgeoning computing technology.

Since the start of the week, IonQ shares have skyrocketed over 50%, as have shares of D-Wave Quantum. Quantum Computing and Rigetti Computing have surged more than 30% each.

The rally comes on the heels of Nvidia’s unveiling of Ising, a new family of open-source models aimed at accelerating the adoption of quantum computing.

“AI is essential to making quantum computing practical,” Nvidia CEO Jensen Huang said in a statement. “With Ising, AI becomes the control plane — the operating system of quantum machines — transforming fragile qubits to scalable and reliable quantum-GPU systems.”

Nvidia explained further in a press release that Ising “provides high-performance, scalable AI tools for quantum error correction and calibration — two of the most critical challenges in building hybrid-quantum classical systems.”

The chip giant named Ising after a famous mathematical model.

Nvidia’s announcement aired on what’s become known as “World Quantum Day,” ever since an international group of scientists announced in 2021 that April 14th should be used to promote public awareness of quantum technology.

The date was chosen because 4.14 represents the first three digits of a key concept in quantum physics known as the Planck constant.

Tuesday also marked Nvidia’s longest winning streak since 2023, with shares up 18% over ten days.

Proponents tout quantum computing as a transformative technology that can accelerate drug discovery and solve problems impossible to answer on everyday computers.

The U.S. government and technology giants are investing heavily in advancing quantum computing.

Over the last few years, hyperscalers such as Microsoft, Alphabet, and Amazon have announced chips to power the futuristic tools. IBM is racing to develop the first large-scale, fault-tolerant quantum computer by 2029.

But the market remains small, with the largest names accounting for around $31 billion in market value ahead of Thursday’s open.

The sector is also susceptible to wide swings due to its speculative nature, and many stocks have slumped year to date. D-Wave and Rigetti have dropped 18% and 12%, respectively.

IonQ also made news on Tuesday. The Maryland-based company said it linked two remote quantum computers, which it called a “foundational technical milestone.” Separately, the company landed a contract with the Defense Advanced Research Projects Agency.

WATCH: De Masi: Quantum systems will be far more energy efficient than classical AI

CORRECTION: This story has been updated to reflect that IBM is developing the first large-scale, fault-tolerant quantum computer by 2029. A previous version of this story misstated that fact.

Three months ago, the tech industry was unaware of a lobster-themed AI coding project built by an under-the-radar Austrian software developer.

OpenClaw, as that creation is known, has enjoyed such a rapid ascent since then that it took center stage this week at GTC, Nvidia’s annual conference, where the leader of the world’s most valuable company called it “the most popular, open-source project in the history of humanity.”

“This is definitely the next ChatGPT,” Nvidia CEO Jensen Huang told CNBC’s Jim Cramer on the sidelines of the developer event in Santa Clara, California. In his keynote, Huang described OpenClaw as the go-to option for building AI agents that can perform tasks like scouting eBay for deals and then placing bids, and said it “exceeded what Linux did in 30 years” in mere weeks.

The phenomenon is so pivotal to Nvidia that the chipmaker said at GTC that it’s building free accompanying security services — packaged as NemoClaw — intended to help spur more adoption of OpenClaw and get large businesses comfortable with its use.

Huang was validating what the rest of the market has been witnessing. An independent developer, rather than a giant, richly valued lab like OpenAI or Anthropic, came up with the next big thing in AI and, in doing so, exposed a potential major flaw in the investment thesis behind the large language models: They may be getting commoditized.

While OpenAI and Anthropic remain deeply popular and continue building services that are resonating with users, the power of OpenClaw is that it’s enabling all sorts of developers and hobbyists to quickly create and manage AI agents across online communications channels like WhatsApp and Telegram from their home computers.

Some industry experts say OpenClaw’s breakout shows that the value in AI isn’t all accruing to the two leading startups, which have a combined private market value of over $1 trillion, and their hyperscaler peers.

“It solidified the open-source community and proved that fully autonomous AI can be run at home without relying on the Magnificent 7 or Big AI,” said David Hendrickson, CEO of consulting firm GenerAIte Solutions. “I suspect this was the black swan moment most big AI companies feared.”

Hendrickson said developers have been gravitating to the Chinese AI models because they are good enough and cheaper to run than the powerful proprietary models from the likes of OpenAI, Anthropic and Google. And because developers use OpenClaw on their personal computers like Apple Mac Minis to manage their fleets of always-operating AI agents, they’ve discovered it’s far more economical than tapping the cloud to access the bigger models.

“As foundation models rapidly commoditize, attention is moving toward agent frameworks that emphasize autonomy, usability, locality, and control to power agentic AI applications and drive business values,” said Charlie Dai, an analyst at Forrester.

OpenAI and Anthropic are well aware of the threat.

Anthropic has been debuting similar OpenClaw-like features, such as a new channels tool.

And last month, in a Sunday post on X, OpenAI CEO Sam Altman announced that Peter Steinberger, the developer of OpenClaw, was joining the AI company and that the service he created would “live in a foundation as an open source project that OpenAI will continue to support.”

Altman called Steinberger a “genius with a lot of amazing ideas,” and said he would help “drive the next generation of personal agents.”

Cohen and his brother have since shuttered their AI marketing firm, created a new startup called NanoCo that will offer paid services to accompany NanoClaw, and partnered last week with container technology company Docker to solidify itself as an OpenClaw competitor.

David Bader, director of the Institute for Data Science at the New Jersey Institute of Technology, said the tech industry is “witnessing a classic platform shift,” with foundation models and Chinese labs “converging in capability.”

“The models become the engine; the agent framework becomes the car,” Bader said.  

Representatives from OpenAI and Anthropic didn’t provide a comment for this story.

Not everyone in the tech industry is convinced that foundation models are losing steam.

Venture capitalist Jerry Chen of Greylock, an Anthropic investor, said OpenClaw’s success in showing what a world of “intelligent agents” can look like doesn’t take away from the importance of the underlying foundation models, which he still sees as more powerful than the so-called open-weight alternatives.

“The buzz around OpenClaw stems from making AI more tangible to a broader audience beyond researchers and technologists,” Chen said. “The interesting question now is whether OpenClaw becomes the de facto standard — the Linux of the market, as Jensen puts it — or just the first of many open and closed-source agentic operating systems.”

For a Wall Street analyst covering Nvidia, the OpenClaw moment is historic in its gravity.

Jay Goldberg of Seaport Research Partners is the lone Nvidia analyst among roughly 70 tracked by FactSet with a sell recommendation on the stock. He initiated his coverage in April after the stock had already rocketed from the AI boom, but the shares kept rallying and are up more than 60% since his sell rating.

“Part of my critique of Nvidia has always been like, what’s the point of all this AI? There’s no consumer use cases for any of it,” Goldberg said. “I’ve always couched my rating by saying, look, where I could be wrong is if somebody comes up with a really incredible AI application.”

After playing around with OpenClaw on a recently purchased Mac Mini, Goldberg said he can finally understand the excitement.

As a parent of three kids, Goldberg said he gets an average of 10 emails a week that he dreads reading, and would love for an agent to scan the messages and tell him of the important stuff like if he has to pick up his kids early from school or get them dressed up for picture day.

“It’s not just the functionality of the thing itself, but it’s the pieces of our lives that we give it access to,” Goldberg said.

Goldberg isn’t ready to boost his rating on Nvidia, but he admitted that he’s “envious” of Huang, who he says “nailed it” in describing OpenClaw as an operating system. Meanwhile, Goldberg said he’s watching tons of TikTok videos on OpenClaw and wants to understand it better before he can feel safe enough to really bake it into his life.

“It’s janky, it is incredibly insecure and it’s like my Mac Mini is kind of half working,” Goldberg said about OpenClaw’s growing pains. “It’s very easy to see how this can become really powerful and really useful.”

WATCH: Nvidia’s one of the fastest growing companies with one of the lowest valuations.

While U.S. markets have been focused on the impact of Anthropic and Altruist’s tools on software and financial services, China’s tech giants have released AI models this week that have shown advancements in robotics and video generation.

Alibaba, TikTok creator ByteDance and short-video platform Kuaishou, have all released new AI models that underscore how Chinese firms are keeping up with those in the U.S.

It comes after Google DeepMind boss Demis Hassabis told CNBC that Chinese AI models are just “months” behind Western rivals.

These models from China are directly competing with video generation models such as OpenAI’s Sora, as well as robotics models from Nvidia and Google.

Here’s a rundown of the models.

However, while users have praised the technology, Seedance has run into trouble. Local Chinese media reported that Seedance has suspended a feature that allowed the AI to generate the voice of a person based on a picture they uploaded. It came after a blogger in China raised concerns about the voice generation taking place without consent.

ByteDance was not immediately available for comment when contacted by CNBC.

The company said the model approaches Anthropic’s Claude Opus 4.5 in coding benchmarks while surpassing Google’s Gemini 3 Pro on some tests. CNBC could not verify those claims.

Shares of MiniMax also jumped Thursday after it launched its updated M2.5 open-source model with enhanced AI agent tools. “Agents” or “agentic AI” refers to AI tools designed to automate tasks.

— CNBC’s Anniek Bao and Dylan Butts contributed to this report.

BEIJING — Chinese companies are accelerating the rollout of new artificial intelligence models as competition with U.S.-based rivals OpenAI, Anthropic and Google intensifies.

Just over a year ago, China-made DeepSeek rocked global markets with the release of an AI chatbot that undercut OpenAI’s ChatGPT on usage fees and production costs, raising questions about the efficacy of U.S. tech restrictions on China.

On Tuesday, Beijing-based startup Moonshot AI revealed Kimi K2.5, which claimed to have video-generation and agentic capabilities that outperformed all three of the leading U.S. AI models. Agentic AI broadly refers to AI systems capable of carrying out tasks on behalf of people. The ultimate goal is to have sophisticated agents that work autonomously with minimal user interaction.

The update came just about three months after Moonshot released its K2 model.

Hours earlier, e-commerce giant Alibaba announced its latest generative AI model, which can create text, pictures or video based on user commands. Alibaba claimed its Qwen3-Max-Thinking outperformed major U.S. rivals on a broad benchmark test called “Humanity’s Last Exam.”

Alibaba said the new model can automatically select the best AI tool for a range of tasks and draw on past conversations as context to generate new responses more efficiently, all at little additional cost.

A week earlier on Jan. 19, Z.ai released a free version of its recently launched GLM 4.7 model. Two days later, the company restricted new subscribers from signing up for its AI coding tool after demand strained its available computing power.

Also last week, Hong Kong-traded shares of Chinese tech giant Baidu climbed to their highest level in nearly three years following the release of its latest generative AI model, Ernie 5.0. The company claimed the update outperformed Google’s Gemini-2.5-Pro, but did not compare it with Google DeepMind’s newest Gemini 3 Pro.

Demis Hassabis, the CEO of Google DeepMind, told CNBC this month that China’s AI models may be just “months” behind those developed in the U.S.

There are indications that the trend is already happening. Microsoft earlier this month noted estimates that DeepSeek usage in Africa is two to four times higher than in other regions.

Paying too much attention to AI benchmarks distracts from the tech’s real value — when it is integrated into existing gaming or entertainment ecosystems such as Tencent’s, Ivan Su, senior equity analyst at Morningstar, said in a note.

The Chinese tech giant operates the ubiquitous WeChat messaging app as well as other popular gaming and video streaming apps in China.

Tencent on Sunday announced that it will distribute 1 billion yuan ($140 million) in cash awards through its Yuanbao AI chatbot app during the Lunar New Year festival in February. The promotion mimics earlier “red envelope” campaigns that helped WeChat become one of the two dominant mobile payment apps just over a decade ago. It is customary for families to give relatives cash in small red envelopes during the holiday.

ByteDance and Baidu are also running AI-related promotions around the upcoming holiday in a bid to retain users on their AI applications.

Earlier this month, Alibaba updated its Qwen AI app to encourage users interacting with the chatbot to shop, order food and pay without leaving the app — thanks to integration with the company’s existing e-commerce platforms such as Taobao. Qwen claims more than 100 million monthly active users.

This integration could generate more revenue for Alibaba, on the assumption that the greater the number of people who use Qwen, the more they will use Taobao and shop on it, Lu said. That would help offset the costs of developing and running the underlying AI model, he said.

Chinese companies are mostly competing for user traffic rather than developing the most advanced models, he said.