{"id":2556,"date":"2026-04-23T18:11:07","date_gmt":"2026-04-23T18:11:07","guid":{"rendered":"https:\/\/deepinsightai.io\/?p=2556"},"modified":"2026-04-23T18:11:08","modified_gmt":"2026-04-23T18:11:08","slug":"hy3-preview-launch","status":"publish","type":"post","link":"https:\/\/deepinsightai.io\/de\/hy3-preview-launch\/","title":{"rendered":"Hy3 Preview by Tencent: Real-World AI Breakthrough in Agents, Coding & Reasoning"},"content":{"rendered":"

Just now, Yao Shunyu led the team to \u201crebuild\u201d Hunyuan, and the first large model has gone live.<\/p>\n\n\n\n

Just now, Tencent Hy3 preview<\/strong> has arrived.<\/p>\n\n\n\n

This is the first version of Hy3<\/strong> released after Tencent\u2019s Hunyuan team restarted from architecture and infrastructure. The initial batch of models is relatively small in size, positioned toward practicality.<\/p>\n\n\n\n

It\u2019s also worth noting that Hy3 preview<\/strong> is the first major result after Yao Shunyu returned to China and joined Tencent. It follows his idea of the \u201csecond half of AI,\u201d where the Hy3 model<\/strong> is refined and improved within Tencent\u2019s real business and complex scenarios, focusing on effectiveness and practicality in real-world use cases.<\/p>\n\n\n\n

Tencent stated that the new generation Hy3 model<\/strong> enhances capabilities in chatting, coding, agents, mathematical reasoning, instruction following, and context understanding.<\/p>\n\n\n\n

\"Hy3<\/figure>\n\n\n\n
\"Hy3<\/figure>\n\n\n\n

Hy3 Deployment Across Tencent Ecosystem<\/h2>\n\n\n\n

At present, Hy3 preview<\/strong> has already launched across Tencent Cloud, Yuanbao, ima, CodeBuddy, WorkBuddy, QQ, QQ Browser, Tencent Docs, and Tencent Lexiang. It is also being rolled out gradually across major products such as WeChat Official Accounts, Peacekeeper Elite, Tencent News, Tencent Self-selected Stocks, Tencent Customer Service, and WeChat Reading.<\/p>\n\n\n\n

In addition, Hy3 preview supports integration with popular open-source agent tools such as OpenClaw<\/a>, OpenCode, and KiloCode, and has been listed on Tencent Cloud\u2019s large model service platform, TokenHub.<\/p>\n\n\n\n

Before the May Day holiday, global AI teams have been making moves one after another. We\u2019ve already seen Qwen 3.6 Max Preview<\/a>, Moonshot AI\u2019s Kimi 2.6, followed closely by Xiaomi\u2019s MiMo-V2.5-Pro.<\/p>\n\n\n\n

So how capable is Tencent Hy3 preview<\/strong> as a foundational large model? Next, we tested Hy3<\/strong> hands-on.<\/p>\n\n\n\n

Hy3 Agent Capability Stress Test: Did It Handle the \u201cLobster Task\u201d?<\/h2>\n\n\n\n

Hy3 Multi-step Reasoning and Research<\/h3>\n\n\n\n

Yao Shunyu is the proposer of the ReAct framework (core logic for agents), so Hy3 preview<\/strong>\u2019s improvements naturally include coding and agent capabilities, aligning with the technical trends and market demand of the agent-driven second half.<\/p>\n\n\n\n

In Tencent\u2019s AI office assistant WorkBuddy, we can ask Hy3<\/strong> to perform coding, in-depth research, product management, data analysis, and more.<\/p>\n\n\n\n

For example, we asked Hy3 preview<\/strong> to investigate rumors about DeepSeek\u2019s financing, requiring it to compare at least five authoritative sources with different backgrounds, list known facts and logical conflicts, and provide confidence scores.<\/p>\n\n\n\n

Hy3<\/strong> was able to autonomously initiate multi-step searches, complete long-chain reasoning, systematically sort out contradictions among sources, and finally present an objective and neutral investigation report\u2014all without human intervention.<\/p>\n\n\n\n

Hy3 Data Analysis and Visualization Tasks<\/h3>\n\n\n\n

We further asked Hy3 preview<\/strong> to fetch data from the United Nations Population Division and complete a visualization analysis on \u201cglobal population structure changes.\u201d<\/p>\n\n\n\n

This is a composite task involving data acquisition, cleaning, analysis, and visualization. Hy3<\/strong> handled it relatively smoothly, ultimately producing intuitive charts and analytical text.<\/p>\n\n\n\n

Hy3 Coding Ability in Practice<\/h3>\n\n\n\n

In coding capability testing, reflecting modern vibe coding<\/a> practices, we asked Hy3 preview to generate a web-based \u201cmatch-three\u201d game.<\/p>\n\n\n\n

The final result had polished visuals, complete logic, and ran normally. The overall quality from Hy3<\/strong> exceeded expectations.<\/p>\n\n\n\n

Hy3 Conversation and Creativity: Are the Fundamentals Solid?<\/h2>\n\n\n\n

We can see that Tencent Hunyuan, like other models, distinguishes between modes: fast thinking for quicker answers, and deep thinking for more thorough reasoning and higher-quality responses. Here, we used deep thinking throughout with Hy3 preview<\/strong>.<\/p>\n\n\n\n

Hy3 Daily Chat and Emotional Understanding<\/h3>\n\n\n\n

This upgrade emphasizes practicality, so we first tried casual conversation with Hy3<\/strong>.<\/p>\n\n\n\n

When we complained to Hy3 preview<\/strong> about feeling less sharp recently, it patiently suggested it might be due to lack of sleep, work pressure, or too much short-video scrolling, and gave three small tips.<\/p>\n\n\n\n

Continuing the topic of lacking inspiration for writing, Hy3<\/strong> naturally connected to the conversation context, adjusted its tone and depth based on our emotional state, and provided targeted creative suggestions.<\/p>\n\n\n\n

It can also deliver strong emotional value, even complimenting in varied ways.<\/p>\n\n\n\n

Hy3 Common Sense Reasoning<\/h3>\n\n\n\n

Previously, Zhihu hosted a discussion called \u201cAI, take this challenge,\u201d which included tricky questions where AI often fails. One question was: \u201cI only found out this year that my biological parents didn\u2019t invite me to their wedding. I feel upset. What should I do?\u201d<\/p>\n\n\n\n

Many large models got confused, ignoring the basic logic that the child wasn\u2019t born yet at the time of the parents\u2019 wedding. Hy3 preview<\/strong> quickly caught this point, guided the user to sort out emotions, and showed strong common-sense reasoning and empathy.<\/p>\n\n\n\n

Hy3 Creative Writing and Style Mimicry<\/h2>\n\n\n\n

Hy3 Social Media Copywriting<\/h3>\n\n\n\n

Recently, a stunning image of NASA astronauts looking back at Earth through the Orion spacecraft window went viral.<\/p>\n\n\n\n

\"a<\/figure>\n\n\n\n

We asked Hy3 preview<\/strong> to generate five social media captions for the image. It first analyzed the atmosphere, chose emotional angles like loneliness and awe, reverence for Earth, and the smallness and greatness of humanity, then produced captions in different styles\u2014some literary, some philosophical. Any one of them from Hy3<\/strong> could be posted directly.<\/p>\n\n\n\n

Hy3 Writing in a Specific Style<\/h3>\n\n\n\n

For style imitation, we asked Hy3<\/strong> to write a short story in the tone of O. Henry.<\/p>\n\n\n\n

Hy3 Search and Information Reliability<\/h3>\n\n\n\n

For search capability, we asked Hy3 preview to investigate why Meta forcibly collects mouse and keyboard input<\/a>. It quickly cited authoritative sources and provided a clear, evidence-based answer.<\/p>\n\n\n\n

Whether checking news, policies, or verifying specific information, Hy3<\/strong>\u2019s overall performance was relatively reliable.<\/p>\n\n\n\n

Hy3 and the Reconstruction of the \u201cSecond Half of AI\u201d<\/h2>\n\n\n\n

According to reports, Hy3 preview<\/strong> is a MoE (Mixture of Experts) language model that integrates fast and slow thinking. The Hy3 model<\/strong> has a total of 295B parameters, with 21B activated parameters, supports a context length of 256K, and balances practicality with cost efficiency.<\/p>\n\n\n\n

In this new generation Hy3<\/strong> model, the Hunyuan team\u2019s main work focused on foundational reconstruction, making many aspects more solid\u2014especially pretraining and reinforcement learning infrastructure, which have been completely rebuilt. Instead of pursuing minor innovations in attention mechanisms or base architecture, the Hy3 architecture<\/strong> follows a mature MoE route and invests heavily in strengthening the engineering infrastructure.<\/p>\n\n\n\n

This means Hy3 preview<\/strong>\u2019s stability, data throughput efficiency, and reinforcement learning pipeline yield may have reached an unprecedented industrial-grade level.<\/p>\n\n\n\n

Hy3 Evaluation Over Training<\/h3>\n\n\n\n

During training, the Hy3 model<\/strong> emphasizes model evaluation and strengthens research on self-built benchmarks. This aligns with Yao Shunyu\u2019s previously shared idea: evaluation is more important than training (Evaluation > Training).<\/p>\n\n\n\n

In \u201cThe Second Half of Large Models,\u201d Yao Shunyu pointed out that the current large model \u201crecipe\u201d (pretraining + reinforcement learning + scaling compute) is already highly mature, capable of generalization and solving complex problems. The key question in the second half becomes: \u201cWhat should we train AI like Hy3<\/strong> to do?\u201d<\/p>\n\n\n\n

Because the general model recipe is already extremely powerful, spending massive effort on fine-tuning may only yield about a 5% improvement. Therefore, evaluation becomes more important than training. The industry needs to reconstruct evaluation systems, designing new tasks and paradigms closer to real-world scenarios, rather than simply making harder test questions.<\/p>\n\n\n\n

Hy3 From Technology to Real-World Problem Solving<\/h2>\n\n\n\n

To survive and grow in the second half of AI, practitioners must shift their mindset and adopt a \u201cproduct manager\u201d perspective. This means deeply thinking about: who should AI like Hy3<\/strong> serve, what real problems it should solve, and how we objectively measure whether it solves them well.<\/p>\n\n\n\n

In this regard, Tencent has some of the most complex business scenarios in China and even globally\u2014covering WeChat, gaming, advertising, and cloud services. Its self-built evaluation environment is naturally highly aligned with real business flows, challenges, and pain points.<\/p>\n\n\n\n

The release of Hy3 preview<\/strong> may already represent a productivity tool within Tencent\u2019s ecosystem that can solve real-world problems.<\/p>\n\n\n\n

Hy3 Is Just the Starting Point<\/h2>\n\n\n\n

Hy3 preview<\/strong> began training at the end of January 2026, and it took less than three months from training to launch. This marks a shift for the Hy3 model<\/strong>\u2014from \u201creading thousands of books\u201d to \u201ctraveling thousands of miles,\u201d starting to tackle real-world problems.<\/p>\n\n\n\n

Hy3 preview<\/strong> is just a starting point. In the future, the Hunyuan team hopes to further improve Hy3<\/strong> capabilities through collaboration with developers and users, allowing the Hy3 ecosystem<\/strong> to continue evolving in real scenarios and tasks.<\/p>\n\n\n\n

<\/p>","protected":false},"excerpt":{"rendered":"

Just now, Yao Shunyu led the team to \u201crebuild\u201d Hunyuan, and the first large model has gone live. Just now, Tencent Hy3 preview has arrived. This is the first version of Hy3 released after Tencent\u2019s Hunyuan team restarted from architecture and infrastructure. The initial batch of models is relatively small in size, positioned toward practicality. […]<\/p>\n","protected":false},"author":1,"featured_media":2559,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_seopress_robots_primary_cat":"none","_seopress_titles_title":"%%post_title%%","_seopress_titles_desc":"Tencent\u2019s Hy3 preview is here. Built for real-world scenarios, the Hy3 model excels in agents, coding, reasoning, and practical AI deployment across Tencent\u2019s ecosystem.","_seopress_robots_index":"","_uag_custom_page_level_css":"","site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[2,10],"tags":[],"class_list":["post-2556","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news","category-llm"],"uagb_featured_image_src":{"full":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch.png",679,454,false],"thumbnail":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch-150x150.png",150,150,true],"medium":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch-300x201.png",300,201,true],"medium_large":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch.png",679,454,false],"large":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch.png",679,454,false],"1536x1536":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch.png",679,454,false],"2048x2048":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch.png",679,454,false],"trp-custom-language-flag":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/hy3-preview-launch-18x12.png",18,12,true]},"uagb_author_info":{"display_name":"Claude Carter","author_link":"https:\/\/deepinsightai.io\/de\/author\/cloud-han03gmail-com\/"},"uagb_comment_info":0,"uagb_excerpt":"Just now, Yao Shunyu led the team to \u201crebuild\u201d Hunyuan, and the first large model has gone live. Just now, Tencent Hy3 preview has arrived. This is the first version of Hy3 released after Tencent\u2019s Hunyuan team restarted from architecture and infrastructure. The initial batch of models is relatively small in size, positioned toward practicality.…","_links":{"self":[{"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/posts\/2556","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/comments?post=2556"}],"version-history":[{"count":1,"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/posts\/2556\/revisions"}],"predecessor-version":[{"id":2563,"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/posts\/2556\/revisions\/2563"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/media\/2559"}],"wp:attachment":[{"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/media?parent=2556"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/categories?post=2556"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/deepinsightai.io\/de\/wp-json\/wp\/v2\/tags?post=2556"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}