On February 18, Elon Musk's artificial intelligence venture, xAI, announced the launch of its latest AI model series called Grok 3. The platform is available to X Premium+ subscribers, who can now access Grok 3's extensive featuresFor those interested in a standalone service, Grok will also offer a membership option called SuperGrok, priced at $30 per month or $300 annually.

Andrej Karpathy, a founding member of OpenAI and former head of Tesla's autopilot division, had the opportunity to test Grok 3 before its official launchHis assessment indicated that Grok 3's capabilities are on par with OpenAI's top-tier model, o1 Pro, which costs $200 per month, and benefits from superior performance compared to other models like DeepSeek-R1 and Gemini 2.0 Flash.

Elon Musk shared that Grok 3 was trained in a massive data center equipped with approximately 200,000 GPUsOne of the standout features of Grok 3 is its introduction of a "Chain of Thought" reasoning ability, which allows the model to tackle complex tasks in a step-by-step manner that closely mirrors human cognitive processesThis enhancement significantly boosts the model's ability to manage intricate inquiries and deliver coherent, logical responsesAdditionally, Grok 3 introduces a new function called DeepSearch, which rivals OpenAI's previously launched deep search, Deep ResearchThis feature enables internet-connected deep searches and summarizes responses for usersNotably, Musk announced that Grok 3 would soon incorporate a voice mode.

A few days later, on February 20, the humanoid robot startup Figure revealed their self-developed AI model, HelixThis new model improves robots' capabilities to handle objects, collaborate with one another, and control upper body motions with precisionThis announcement came shortly after Figure ended its partnership with OpenAI just a few weeks prior, and was quickly followed by news of a $1.5 billion financing round.

According to Figure, the launch of Helix represents a crucial milestone in making humanoid robots suitable for everyday household environments

Advertisements

Unlike earlier models, Helix does not require extensive training through demonstrations to complete tasksIts innovative architecture enables robots to interact with objects they have never previously encounteredIn a showcase video, two Figure robots successfully worked together to organize a grocery bag and place items in the refrigerator, while arranging apples in a fruit bowl.

Previously, Figure had collaborated with OpenAI to customize an AI model for its robotsHowever, on February 4, Figure's founder Brett Adcock announced on social media that they had terminated that collaboration, stating that Figure achieved a breakthrough in developing a "fully end-to-end, fully autonomous" robot AI, which refers to HelixAdcock disclosed that the company had spent over a year developing Helix.

The company's latest progress has drawn significant interest from the investment communityFigure is currently in talks with Align Venture and Parkway Venture Capital for a new funding round, which is expected to raise $1.5 billion, boosting the company's valuation to $39.5 billionThe enthusiasm from investors stems from Figure's accelerated efforts to bring humanoid robots into the home market.

The humanoid robot sector is witnessing intensified competition, with Meta recently establishing a dedicated department for researching AI-driven humanoid robotsStartups like Apptronik have secured $350 million in funding, with investors that include Google DeepMindTesla is also advancing its "Optimus" robot project, while rumors suggest that Apple might be entering the fray as well.

Meanwhile, Tencent's suite of products has started integrating DeepSeek after enhancing its WeChat search featuresOn February 15, WeChat initiated gray testing for its new "AI search" feature, incorporating DeepSeek's R1 model to enhance its reasoning capabilitiesTencent claims users can experience the full version of DeepSeek R1, equating to the original model's 671 billion parameters, which supports online searching and consolidates data from Tencent's ecosystem.

Users can access AI search through a search entry at the top of their WeChat chat window, offering quick and deeply reasoned responses

Advertisements

Though currently undergoing gray testing—accessible to a limited number of users—Tencent has provided an option for those who missed the opportunity to download the "Tencent Yuanbao" app to utilize the full R1 model as well as the DeepSeek-V3 model for free.

Tencent's Yuanbao, set to launch in May 2024, features an AI assistant designed for writing, image generation, and other functionalitiesFollowing updates on February 13, Yuanbao integrated two significant models, Hunyuan and DeepSeekBy February 17, the app announced not only support for DeepSeek R1 but was also beginning gray testing for its in-house Hunyuan T1 model, touted as a deep thinking model.

Moreover, numerous Tencent products have embraced DeepSeek technologyFor example, the AI document assistant in Tencent Docs is paired with the full version of DeepSeek-R1 to generate documents, spreadsheets, presentations, and mind mapsFurthermore, QQ Music has integrated DeepSeek-R1 for improved music recommendations, knowledge Q&A, and celebrity news, among other applicationsAdditional offerings such as the AI smart workspace ima.copilot, QQ Browser, Tencent Maps, and more are leveraging this technology as well.

Not to be outdone, Baidu's mapping and search features also engaged with DeepSeekOn February 15, Baidu Maps announced its full integration with DeepSeek-R1, allowing users to inquire about various activities, with DeepSeek considering past search history to provide comprehensive responses that include business names and specific addresses with user reviews.

The very next day, Baidu confirmed its search engine would implement DeepSeek alongside its Wenxin large model to deliver profound searching capabilitiesIn parallel, the Wenxin intelligent platform declared plans to integrate DeepSeek for developersPresently, Baidu's search service has not yet applied DeepSeek, though the intelligent platform is running trials and is anticipated to enable DeepSeek utilization soon.

In the vast technological ecosystem, Alibaba announced its objective to invest more in AI over the next three years than it had in the past decade

Advertisements

On February 20, the company revealed its financial results for the third quarter of the 2025 fiscal year, citing total revenue of 280.15 billion yuan, marking an 8% increase year-on-year, with operating profit growing by 83% to 41.205 billion yuan.

From its diverse sectors, the biggest income source comes from the Taotian Group with a revenue of 136.09 billion yuan, which reflects a year-on-year growth of 5%. The Cloud Intelligence Group reported revenue of 31.742 billion yuan, experiencing a 13% increase, while AI-related products saw consistent triple-digit growth over six consecutive quartersThe International Digital Commerce Group achieved 37.756 billion yuan, showing 32% growth, greatly influenced by robust cross-border business performanceThe Local Life Group accrued 16.988 billion yuan, growing 12%, driven by increases in orders from Gaode and Ele.meHowever, the Cainiao Group observed a 1% drop in revenue, whereas the Digital Media and Entertainment Group grew by 8%.

During the financial conference call, Alibaba's CEO Wu Yongming expressed plans for substantial investment in AI across several areas such as AI infrastructure, foundational model platforms, AI-native applications, and AI transformations of existing businessesParticularly, investments in AI infrastructure are poised to exceed the total funding from the past decadeWu also indicated that Alibaba would launch a depth inference model based on Qwen2.5-Max soon.

The Cloud Intelligence Group experienced growth in the first, second, and third quarters of 2025, with increases of 6%, 7%, and 13%, respectively, marking an all-time highAnalysts are optimistic about Alibaba's cloud business significantly outperforming Wall Street expectations, as its AI strategy appears to be gaining momentumFollowing the earnings report, Alibaba's stock soared by more than 10% in pre-market trading, rising over 60% since the beginning of the year.

In a significant reshuffle, ByteDance has undergone a transformation in its AI teams, appointing Wu Yonghui as a co-leader of the Seed foundational model development division

This comes alongside Zhu Wenjia, maintaining an egalitarian structure with both reporting directly to CEO Liang RuboWu's role focuses on the "foundational models" aspect while Zhu concentrates on "model applications."

Previously, Zhu had singular authority over the Seed team, having formed it at the beginning of 2023 while serving as the CEO of Toutiao and the head of technological development for TikTokAlthough various teams within ByteDance were exploring key underlying technologies, Zhu's Seed team emerged as the principal success story, cementing his pivotal role in the AI sector.

With Wu's elevation to coordinate all foundational model efforts, Zhu's areas of responsibility have drastically narrowedDetails surrounding the scope of Zhu's focus on "model applications" remain undetermined, especially considering that ByteDance has a dedicated team, Flow, already focused on AI innovative applicationsPreviously, the divisions between Seed and Flow were clearer, as the AI assistant Doubao was developed by Flow, overseen by its leader Zhu Jun, a co-founder of Musical.ly, the predecessor of TikTok.

Ultimately, Wu's rise represents a substantial victory in the leadership reshuffleRecent reports emerged that he joined ByteDance on February 17, initially announced to lead the Seed Edge team, a new unit launched within Seed, which aims at long-term foundational AGI research and next-generation inference systems.

However, it has now become apparent that Wu will head all model development within SeedSeveral algorithmic and technical leaders who previously reported to Zhu are now directed by Wu, including heads from ByteDance's AML and Doubao large model Foundation teams, as well as the directors of ByteDance’s artificial intelligence labs and related visual multimodal generation teams.

In the wake of these developments, Tencent made headlines by transferring several of its applications—including QQ Browser, Sogou Input Method, and Ima—to its Cloud and Smart Industries Group (CSIG). This maneuver indicates a further alignment of high-performing products closely associated with AI large models.

Following a prior shift in January that moved the AI assistant Yuanbao from the TEG structure to CSIG, the now-transferred products each have established user bases and substantial content, ideally placing them for competitive advantages in AI search and semantic comprehension.

In an industry that never ceases to evolve, collaboration, competition, and technological advancements remain at the forefront, setting the stage for continuous transformation and innovation in not just AI capabilities, but in the very fabric of day-to-day digital interaction.

Advertisements

Advertisements