Tencent announced a broader international rollout of Hy3, formerly known as Hunyuan, on August 5, 2026. The model became available through three Tencent products—WorkBuddy, Tencent Design Miora, and Tencent Cloud’s TokenHub—while developers could also connect it to applications through APIs and third-party tools. WorkBuddy access was offered free worldwide through August 31, 2026, Pacific Time.
2
5
6
20
The announcement was less about unveiling a new model than about expanding the number of ways people could use Hy3: as a workplace agent, a creative assistant, or a model accessed through cloud infrastructure and developer platforms.
Where global users could access Hy3
Tencent’s rollout created three primary entry points:
- WorkBuddy: An agentic workplace environment for general users and enterprises.
- Tencent Design Miora: A creative and design-oriented application for working with the model.
- Tencent Cloud TokenHub: A model-access and deployment platform for developers and businesses.
Developers could also connect Hy3 through APIs to development environments, coding extensions, external platforms, and other tools. Reports said integrations were available or expanding across platforms including OpenRouter, Hermes Agent, Kilo Code, Claude Code, Hermes, Kilo, Cline, OpenClaw, OpenCode, and Cherry Studio.
2
7
9
10
That distribution strategy gave Hy3 different paths into the market. WorkBuddy targeted everyday and workplace use, Miora focused on creative production, and TokenHub provided a cloud-oriented route for organizations building applications around models.
WorkBuddy’s free-access offer
Tencent said users worldwide could try Hy3 in the international version of WorkBuddy without charge until August 31, 2026, Pacific Time. Because that date has passed, the offer should be understood as a limited launch promotion rather than an ongoing free-access policy.
1
5
6
11
14
15
The promotion also served as a way to reduce friction for early users. Instead of requiring developers or companies to begin with an API integration, Tencent gave them a direct way to test Hy3 inside a ready-made agent workspace.
Hy3’s release and technical profile
Tencent had officially released Hy3 on July 6, before the international expansion. Reports described the model as available under the Apache 2.0 license on repositories including Hugging Face, ModelScope, GitHub, and GitCode.
3
9
10
The reported specifications were:
- 295 billion total parameters
- 21 billion active parameters
- A context window of up to 256K tokens
- A hybrid “fast-and-slow-thinking” mixture-of-experts architecture
Tencent claimed Hy3 could perform competitively with models containing two to five times as many parameters. That is a company claim, not independent verification, so the parameter comparison should not be treated as a neutral benchmark result.
9
10
15
The model’s design was presented as suited to reasoning and agentic tasks, while its availability through coding tools and APIs broadened its potential use beyond Tencent’s own products.
Usage and OpenRouter ranking claims need context
Tencent said Hy3’s API call volume had increased by more than 68 times compared with its previous-generation model and that Hy3 had briefly reached the top of OpenRouter’s global usage leaderboard.
3
5
6
7
9
10
The ranking picture reported on August 5 was more nuanced. One weekly OpenRouter ranking placed Hy3 third, behind DeepSeek V4 Flash and Xiaomi MiMo-V2.5.
2
34
36
37
These statements are not necessarily contradictory: a model can briefly lead a live or different measurement window while ranking third in a weekly snapshot. OpenRouter rankings also measure routed token usage over defined time windows, so positions can change rapidly. Later reporting showed Hy3 still near the top but behind other models in some weekly measurements.
1
3
33
35
The practical takeaway is that Hy3 showed strong early adoption signals, but a temporary leaderboard position is not the same as a stable global ranking or proof of superior model quality.
Reported API pricing
The supplied reporting cited OpenRouter pricing of $0.1288 per million input tokens and $0.5336 per million output tokens for Hy3.
1
11
Those figures should be treated as a time-specific listing rather than a permanent price. Model aggregators can change prices, and the available material does not independently verify that the quoted rates applied across every Hy3 access route, including TokenHub or Tencent’s own products.
Why the expansion matters
The international rollout reflected Tencent’s attempt to make model access part of a broader deployment stack instead of treating Hy3 as a standalone open-source release. The company connected the model to workplace agents, creative software, APIs, coding tools, and cloud infrastructure in one launch.
5
6
13
20
The South China Morning Post framed the move as part of a wider push by Chinese technology companies to compete more directly in the global AI market.
5
Tencent Cloud International executive Poshu Yeung described the company’s stated objective as helping enterprises move from AI experimentation toward secure, reliable, and scalable production deployment through models, applications, orchestration, and enterprise infrastructure. That is Tencent’s strategic positioning, not independent evidence that every part of the deployment stack has been validated in production.
6
The bottom line
Tencent’s August 5 announcement made Hy3 globally easier to try and integrate. The key elements were:
- International availability through WorkBuddy, Tencent Design Miora, and Tencent Cloud TokenHub.
- Free worldwide WorkBuddy access through August 31, 2026, Pacific Time.
- API access for development environments, coding extensions, and third-party platforms.
- An Apache 2.0 release with reported 295-billion total parameters, 21-billion active parameters, and a 256K context window.
- Strong early usage claims, balanced by volatile OpenRouter rankings and the lack of independent verification for Tencent’s performance comparisons.
Hy3’s significance therefore lies as much in its distribution as in its architecture: Tencent was putting one model across consumer workflows, developer ecosystems, and enterprise cloud deployment at the same time.