AnswersPublished20 sources
Tencent Hy4 Adds Vision and Computer Use, but Its Open-Weight Ranking Needs Context
Tencent’s November 4 Hy4 release adds visual understanding and computer use; reported results include 83.18 in WorkBuddy and 3.04/4 on finance tasks. Hy4’s clearest reported changes are stronger long horizon and instruction following performance, plus the ability to work with images and software interfaces.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Tencent’s November 4 Hy4 release adds visual understanding and computer use; reported results include 83.18 in WorkBuddy and 3.04/4 on finance tasks.
- Hy4’s clearest reported changes are stronger long horizon and instruction following performance, plus the ability to work with images and software interfaces.
- Tencent reports first place among open weight models on its internal TenureTrail research benchmark, but the source material gives no numerical score for that result.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What does Tencent’s November 4, 2026 release of Hy4 reveal about its improvements over Hy4 preview and standing among top open-weight models. Article summary: Tencent’s November 4 release presents Hy4 as a substantial upgrade from Hy4 preview, adding vision and computer use while improving reasoning and sustained agentic work; Tencent places it among the strongest open-source . Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
Tencent’s November 4, 2026 release positions Hy4 as more than a text model: it adds visual understanding and computer use, while targeting better instruction following, long-context comprehension and sustained agent work. Reported office, finance and research results are encouraging, but they come from specific evaluations; they do not establish a universal ranking against every leading model.3
8
18
20
What changed from Hy4 preview
Tencent says the full release expands pre-training and post-training and improves instruction following, long-context understanding, reasoning efficiency and long-horizon agent tasks.3
8 The most visible additions are visual input and computer use—capabilities that let Hy4 work with images and interact with software rather than only respond to text.
5
Tencent and launch coverage describe Hy4 interpreting charts and documents, reasoning across images, and using command-line tools and graphical interfaces. Tencent also says it trained the model on professional software such as Blender and CAD tools, with an emphasis on complex, extended workflows.5
25
What the reported evaluations show
WorkBuddy: strong results in a defined office test
Tencent reports a WorkBuddy evaluation average of 83.18, a 96.15% task pass rate, and first place in 9 of 12 scenarios in the mainland Chinese market.25 These figures indicate strong performance in that evaluation; they do not establish how reliably Hy4 will handle every company’s workflows or unscripted office tasks.
Finance: a lead in one comparison
In a reported crowdsourced evaluation of 579 real financial tasks, Hy4 scored 3.04 out of 4. The reported comparison scores were 2.97 for DeepSeek-V4.1-Flash-1017 and 2.68 each for Kimi K3.1 Pro and GLM-5.3. In this particular comparison, Hy4 ranked first among the open-source models tested.18
19
Tencent also describes a separate blind evaluation conducted with Tencent Financial Technology.25 The available source details do not provide enough information about that test’s protocol or results to assess it independently. The 3.04 score should therefore be read as a result from the reported 579-task evaluation, not assumed to be the blind-test score.
Scientific research: a first-place claim on an internal benchmark
Tencent says Hy4 ranked first among open-weight models on TenureTrail, its internal benchmark of 123 executable research tasks across seven disciplines.20 The announcement describes tasks spanning areas such as mathematical construction, physical modeling and systems engineering, but the available material does not give a numerical score. This is a company-reported result on an internal benchmark, not a general-purpose measure of research ability.
Does Hy4 rank among the top open-weight models?
Tencent describes Hy4 as among the strongest open-source models, and its reported results support competitiveness in selected office, finance and research evaluations.7
18
20 But benchmark leaders can differ by task, and these results do not establish an independent overall ranking across models.
There is also an important release distinction: Hy4 was announced and made available through Tencent products and hosted API routes, while coverage of the launch said the final model weights and system cards would follow. Hy4 preview, by contrast, had already been released with open weights.3
8
9 So the release announcement’s “open-source” positioning should not be mistaken for confirmation that the final Hy4 weights were already downloadable on November 4.
What remains unclear
The available material does not provide enough detail to verify specific WorkBuddy task examples or enterprise-adoption figures, supported game engines, or detailed front-end and game-development results. It also does not establish Hy4 Flash’s launch timing or pricing, or a roadmap for future model sizes and releases.
Tencent’s Hy4 page describes an internal expert audit of certain offensive cybersecurity capabilities and says it found no prominent risk in the areas examined.21 That is a company-reported finding, not an independent safety assessment; the available source material does not establish biological-safety results.
The clearest takeaway is that Hy4’s release broadens the model toward visually grounded software use and extended professional workflows. Its reported evaluation results are promising, but claims about overall standing should remain tied to the specific tests—and to the distinction between a hosted release and publicly released weights.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Tencent Hy4 Adds Vision and Computer Use, but Its Open-Weight Ranking Needs Context"?
Tencent’s November 4 Hy4 release adds visual understanding and computer use; reported results include 83.18 in WorkBuddy and 3.04/4 on finance tasks.
What are the key points to validate first?
Tencent’s November 4 Hy4 release adds visual understanding and computer use; reported results include 83.18 in WorkBuddy and 3.04/4 on finance tasks. Hy4’s clearest reported changes are stronger long horizon and instruction following performance, plus the ability to work with images and software interfaces.
What should I do next in practice?
Tencent reports first place among open weight models on its internal TenureTrail research benchmark, but the source material gives no numerical score for that result.
Sources
nocode.mbaTencent Hy4: Release Date, Access & What Changed
siliconangle.comTencent releases Hy4, powerful frontier LLM introducing visual understanding and computer use - SiliconANGLE
longbridge.comTencent Hunyuan Officially Releases New-Generation Large Language Model Hy4, Introducing Visual Capabilities for the First Time
hy.tencent.aiIntroducing Hy4 - Tencent Hy