Hopp til hovedinnhold
Lørdag 25. april 2026AI-nyheter, ferdig filtrert for ledere
SISTE:
DeepSeek åpner V4 Preview med 1M kontekst og API-kompatibilitetOpenAI lanserer GPT-5.5 for ChatGPT og CodexAnthropic og Amazon utvider AI-alliansen med 5 GW kapasitet og ny investeringDeepSeek åpner V4 Preview med 1M kontekst og API-kompatibilitetOpenAI lanserer GPT-5.5 for ChatGPT og CodexAnthropic og Amazon utvider AI-alliansen med 5 GW kapasitet og ny investering
Tencent releases Hy3 Preview with 256K context and an open model strategy
CIOOpen SourceAI Strategy

Tencent releases Hy3 Preview with 256K context and an open model strategy

JH
Joachim Høgby
25. april 20263t siden4 min lesingKilde:

Tencent has released Hy3 Preview, a new open Hunyuan model and the company’s first major model update since former OpenAI researcher Yao Shunyu took a central role in Tencent’s foundation-model effort.

For executives, this is not just another China model launch. It shows Chinese platforms moving quickly toward practical enterprise deployment: fewer active parameters, longer context, stronger agent capabilities, and tighter integration into their own products.

What is new

According to Tencent’s Hy3 page and coverage from SCMP, Caixin and Computerworld, Hy3 Preview is a mixture-of-experts model with 295 billion total parameters, with 21 billion active per task. The model supports up to 256,000 tokens of context.

Tencent says the model is already deployed across its own ecosystem, including Yuanbao, CodeBuddy and Tencent Docs. That matters. Many model launches stop at a benchmark table. Tencent is trying to turn Hy3 into an actual platform component inside products it already distributes.

Why it matters

This increases pressure on both price and distribution in the enterprise AI market. Tencent is not only competing model-to-model on benchmarks. It is competing with a Chinese cloud, app and agent stack.

For CIOs, there are three takeaways. First, model strategy should stay multi-model, not vendor-religious. Second, regional requirements will matter more, especially for companies with operations or supply chains in Asia. Third, open models from large platform companies are no longer just a developer story. They can become leverage in procurement, cost control and sovereignty discussions.

That does not mean European companies should move sensitive workflows to Tencent without scrutiny. It means they should pay attention. When Chinese vendors combine open models, lower active compute and distribution inside major work tools, the competitive pressure also reaches Western providers.

Source and age validation

Tencent’s official Hy3 Research page is available at hy.tencent.com/research/hy3, with build time 2026/4/23 23:51:37. SCMP, Caixin Global and Computerworld published coverage on April 24, 2026 and confirm that Hy3 Preview has launched/is available in preview. This is within the 48-hour requirement.

Sources:

  • https://hy.tencent.com/research/hy3
  • https://www.scmp.com/tech/big-tech/article/3351101/tencent-unveils-first-flagship-ai-model-former-openai-researcher-helm
  • https://www.caixinglobal.com/2026-04-24/tech-brief-april-24-tencent-unveils-first-major-ai-model-update-under-new-leadership-102437622.html
  • https://www.computerworld.com/article/4163293/former-openai-research-scientist-launches-new-ai-model-for-tencent-2.html

📬 Likte du denne?

AI-nyheter for ledere. Kuratert av en CIO som bygger det selv. Daglig i innboksen.

Relaterte saker

Pony.ai pushes more AI compute into the vehicle
CIOAIInfrastructure

Pony.ai pushes more AI compute into the vehicle

Akkurat nå3 min lesing
Åpne saken
Pony.ai flytter mer AI-compute ut i kjøretøyet
CIOAIInfrastruktur

Pony.ai flytter mer AI-compute ut i kjøretøyet

Akkurat nå3 min lesing
Åpne saken
Tencent slipper Hy3 Preview med 256K kontekst og åpen modellstrategi
CIOOpen SourceAI Strategy

Tencent slipper Hy3 Preview med 256K kontekst og åpen modellstrategi

3t siden4 min lesing
Åpne saken