Tencent opens Hy3 globally via WorkBuddy, Miora and TokenHub
- Source
- x.com
- Date
Tencent Hy3 is now available globally through WorkBuddy, Tencent Design Miora and Tencent Cloud TokenHub. Designed to be compact and cost-efficient while delivering flagship-level performance, Hy3 brings advanced reasoning and agent capabilities to users, developers and enterprises. Try Hy3 today. Hy3 Official Website: https://t.co/iHA8HFYqQV Hy3 API Pricing: https://t.co/e1cQYkzemN Tencent WorkBuddy: https://t.co/MOL7DW7sQw Tencent Design Miora: https://t.co/OuHlFO6cWu Tencent Cloud TokenHub:
Context
Tencent's August 5 post is about access rather than a new model. Hy3, which Tencent released on July 6 as open weights under the Apache 2.0 license, could already be downloaded and run by anyone with the hardware. This announcement puts it inside three hosted Tencent products that international users can sign up for. According to Tencent's launch announcement, Hy3 is a model, a design that holds 295 billion parameters but uses only 21 billion of them for each token, with a of 256K tokens.
The three routes serve different users. WorkBuddy is an workspace for research, data analysis, documents, presentations and office messages, with more than 100 built-in skills and remote control through Discord, Slack and Telegram; Tencent made it free for international users until August 31, 2026. Miora is a design workspace that coordinates several specialized agents to turn a plain-language request into graphics, video, 3D content or interface designs. TokenHub is Tencent Cloud's developer gateway, one API for Tencent's own Hy models and third-party models. Its product page, which now also lists the later Hy4 preview, names DeepSeek, GLM, Kimi, MiniMax and Qwen among those third-party models and offers pay-as-you-go billing, guaranteed resources or dedicated deployment.
The performance claims are Tencent's own. The company says Hy3 rivals flagship models two to five times its size and that, in internal tests, WorkBuddy running on Hy3 completed more than 90% of tasks and finished them in 34% less time than with the previous Hy model, but it does not name the tasks or comparison models. The one concrete price it gives is on OpenRouter, from US$0.1288 per million input tokens and US$0.5336 per million output tokens. For anyone choosing between these routes, the practical difference is that the hosted products remove the need to serve a 295-billion-parameter model yourself, at the cost of relying on a hosted service and on performance claims Tencent has not shown how to check.
- open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
- mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…







