/
Navigation
Chronicles
Browse all articles
Explore
Semantic exploration
Research
Entity momentum
Nexus
Correlations & relationships
Story Arc
Topic evolution
Drift Map
Semantic trajectory animation
Posts
Analysis & commentary
Pulse API
Tech news intelligence API
Browse
Entities
Companies, people, products, technologies
Domains
Browse by publication source
Handles
Browse by social media handle
Detection
Concept Search
Semantic similarity search
High Impact Stories
Top coverage by position
Sentiment Analysis
Positive/negative coverage
Anomaly Detection
Unusual coverage patterns
Analysis
Rivalry Report
Compare two entities head-to-head
Semantic Pivots
Narrative discontinuities
Crisis Response
Event recovery patterns
Connected
Search: /
Command: ⌘K
Embeddings: large
TEXXR

Chronicles

The story behind the story

days · browse · Enter similar · o open

Tencent releases Hy3, a 295B-parameter model that it says is competitive with GLM-5.1 and 5.2, under the Apache 2.0 license, following a preview launch in April

For the past year, the awkward secret of the open-weight model boom has been that many of the strongest Chinese releases …

VentureBeat Sam Witteveen

Context & Ripple Effects

Tencent’s April Hy3 preview marked a new model effort under former OpenAI researcher Yao Shunyu, with a 295B-parameter design that was smaller than the company’s prior HY2. The full Hy3 release turns that preview into an openly licensed offering.

The release arrives shortly after Z.ai positioned its much larger GLM-5.1 open-weight model as a high-performing alternative on software-engineering evaluation. Tencent is now explicitly placing Hy3 in that same competitive set, while using the permissive Apache 2.0 license rather than a restricted distribution model.

First-order effects

  • Developers and enterprises can now obtain, modify, and deploy Tencent’s 295B-parameter Hy3 under Apache 2.0 terms, reducing licensing friction for adoption and downstream customization.
  • Tencent becomes a more direct open-model competitor to Z.ai’s GLM line, with its performance claim inviting immediate comparison on practical workloads rather than parameter count alone.

Second-order effects

  • Z.ai and other open-weight model vendors face added pressure to substantiate benchmark claims and differentiate through efficiency, tooling, multimodal capabilities, or deployment support—not just larger model scale.
  • Permissive licensing gives cloud providers, model-hosting platforms, and enterprise AI teams another large Chinese model to evaluate for self-hosted stacks, potentially broadening the supplier set available to them.

Third-order effects

  • If comparable models continue to be released under permissive terms, the competitive center of open-weight AI may shift further from access to raw weights toward evaluation credibility, inference efficiency, and the ecosystems built around deployment.
  • The Hy3-versus-GLM comparison also suggests that parameter count is becoming a less sufficient proxy for competitive position; whether that holds depends on independent validation and real-world operating costs.

The trend: This is another step in the rapid commercialization of permissively licensed, high-capability Chinese open-weight models competing on usable performance rather than exclusivity.

Discussion

  • @tencenthunyuan Tencent Hy on x
    🚀Hy3 is here. 295B MoE. Best in its size class. Rivals trillion-scale flagships. Reliable and affordable for most agentic usecases. Apache 2.0. Friendly for commercial use. FREE API for 2 weeks → https://openrouter.ai/... 🤗 https://huggingface.co/... 📖 https://hy.tencent.com/... …
  • @rogerw0108 Roger Wang on x
    🔥Huge congrats to the @TencentHunyuan team for the incredible model release! Alongside Hy3, we've been working closely with the team to bring their HPC-Ops natively into vLLM! See full technical details in our blog post and take Hy3 for a spin today🚀 https://vllm.ai/...
  • @gneubig Graham Neubig on x
    Interesting new release Hy3. If the numbers are really reflected in real-world vibes, this is a step closer to a strong model you can host yourself (more easily than GLM-5.2).
  • @shunyuyao12 Shunyu Yao on x
    Hy2 -> Hy3 preview -> Hy3 Another massive leap forward, under half a year. Not just a leap of reasoning or agentic capabilities. Also a leap of anti-hallucination, reliability, and product experiences. More on the way and so proud of the team! 🧑‍🍳🧑‍🍳🧑‍🍳 [image]
  • @tencentai_news @tencentai_news on x
    Introducing Hy3, our latest Hy model🙌 It outperforms similar-sized models and rivals flagship open-source models with 2 to 5× the parameters. — Strong, practical gains in coding, office productivity, financial modeling, frontend design, and game development — Blind-tested by
  • @zixuanli_ Zixuan Li on x
    Hy3 from @TencentHunyuan is out. Great to see real-world workflows emphasized, with GLM-5.1 cited in the comparison: “we ran a blind test with 270 experts from various disciplines, working on real-world workflows. Hy3 scored 2.67/4, outperforming GLM-5.1 at 2.51/4.” [image]
  • @jenzhuscott Jen Zhu on x
    So @TencentHunyuan @ShunyuYao12 did not disappoint - just dropped Hy3. Apache 2.0. The numbers 🔥🔥 - 295B total params, 21B active. Compare to DeepSeek/Qwen at ~1T, overseas frontier at ~10T. - API: ¥1 / ¥4 / ¥0.25 per M tokens. Cheapest Chinese model on the market. Cheaper [image…