Tencent
Hy3
hy3
Total Context
256K
Max Output
128K
Released
N/A
Input$0.1429 / M tokens
Output$0.5714 / M tokens
Hy4 preview
hy4-previewHy4 preview is Tencent Hunyuan’s next-generation open-source Mixture-of-Experts flagship model, built for real-world productivity across coding, office work, scientific research, long-context reasoning, and complex agent workflows. Also known as Tencent Hy4 preview or Hunyuan 4 preview, it has 770B total parameters, activates 49B per token, and supports a context window of over 1M tokens. It excels at understanding large codebases and documents, multi-step planning, tool use, debugging, validation, and sustained task execution.
Context Window
1M tokens
Maximum Output
64K tokens
Release Date
Aug 28, 2026
Modalities
Hy4 preview Pricing
| Input Price | Output Price | Cache Read |
|---|---|---|
| $0.8571/M | $2.5714/M | $0.0429/M |
Hy4 preview API Capabilities
Reasoning
Supported
Tool calling
Supported
Temperature parameter
Supported
Attachments
Not supported
Knowledge Base
—
Endpoint Protocols
Completions APIResponses APIMessages API
Hy4 Preview Model Highlights
Hy4 Preview combines long-horizon engineering, productivity and scientific reasoning, and a one-million-token context in a text-focused flagship model.
Long-Horizon Engineering
The model can understand, plan, debug, and verify extended development tasks while retaining continuity across multiple steps.
Productivity Reasoning
Hy4 Preview works across office analysis, financial models, multi-file information, and structured professional deliverables.
Scientific Problem Solving
The model supports reasoning over difficult questions in AI research, molecular dynamics, condensed-matter physics, and mathematics.
Hy4 Preview Use Cases
Hy4 Preview is suited to complex software development, cross-document office analysis, game prototyping, and technical research.
End-to-End Software Development
Analyze requirements, plan implementation, modify a large codebase, diagnose failures, and verify the completed solution.
Cross-Document Analysis
Consolidate scattered files and data into reports, spreadsheets, presentations, financial models, or other shareable artifacts.
Scientific Research Analysis
Work through technical literature and difficult scientific questions to derive explanations, compare approaches, and formulate research directions.
How to Use Hy4 preview via the TokenHub API
import OpenAI from "openai"
const client = new OpenAI({
apiKey: process.env.TOKENHUB_API_KEY,
baseURL: "https://us-api.tokenhub.com/v1",
})
const result = await client.chat.completions.create({})
console.log(result.choices[0]?.message?.content)Hy4 Preview FAQs
Answers about Tencent Hunyuan Hy4 Preview, its architecture, long-context capabilities, productivity use cases, tradeoffs, API usage, and model selection.
What is Hy4 Preview?+
Hy4 Preview is Tencent Hunyuan’s next-generation open-source Mixture-of-Experts flagship model. Also known as Tencent Hy4 Preview or Hunyuan 4 Preview, it has 770B total parameters, activates 49B parameters per token, and supports a context window exceeding 1M tokens. It is designed for coding, office work, scientific research, reasoning, and complex agent workflows.
What is Hy4 Preview best suited for?+
Hy4 Preview is well suited to large-repository coding, software planning and debugging, long-document analysis, multi-file office tasks, scientific research, tool-using agents, and workflows that must preserve context across many steps. Its productivity focus makes it most relevant when a task requires sustained execution rather than a single short response.
What are the main strengths of Hy4 Preview?+
Its main strengths are long-context understanding, multi-step planning, coding, tool use, debugging, validation, and sustained agent execution. The sparse MoE architecture provides much more total model capacity than the number of parameters activated for each token, while the open weights make deployment and technical evaluation possible outside a single hosted product.
What tradeoffs should I consider with a preview model?+
Hy4 Preview is an early release rather than the final Hy4 model, so behavior, availability, pricing, and integration details may change. Teams should regression-test important workflows, review tool actions, monitor long reasoning traces, and keep a fallback model for production systems. Large open-weight deployments also require substantial serving resources.
How do I use Hy4 Preview through the TokenHub API?+
Use the Hy4 Preview model ID displayed in the TokenHub catalog with your TokenHub endpoint and API key. Follow the selected endpoint’s format for messages, reasoning options, streaming, tools, and structured output. If you are migrating from a Tencent Hunyuan API integration, validate request fields, tool schemas, and response parsing before moving production traffic.
How should I evaluate Hy4 Preview pricing and availability?+
Check the current TokenHub model page for input, output, cache, or request-based pricing and supported endpoints. Estimate cost with your real prompt length, expected output, reasoning behavior, concurrency, tool calls, and retry rate. Because this is a preview model, avoid treating launch pricing or temporary availability as a permanent service commitment.
Should I choose Hy4 Preview or Hy3?+
Evaluate Hy4 Preview first when you need much longer context, stronger multi-step execution, large-scale coding, document-heavy work, or scientific research. Hy3 may be the safer choice when an existing production workflow already depends on its tested behavior, latency, or integration contract. Run both models on the same representative tasks before migrating.
Ready to use Hy4 preview?
Use one API key to access Hy4 preview and more AI models through TokenHub.
Hy4 Preview Media and Demos
Selected public announcements, technical explainers, deployment tests, and community discussions about Tencent Hunyuan Hy4 Preview.
X (Twitter)
Tencent Hy ·
🚀 Hy4 preview is here. 770B, 49B active, 1M context. Built for productivity. Open source frontier. Consistent affordable price. Use it. Tell us what breaks. More on Hy blog：https://hy.tencent.ai/research/hy4-preview HuggingFace：https://huggingface.co/tencent/Hy4-preview Github：https://github.com/Tencent-Hunyuan/Hy4-preview https://t.co/4iW6gSuZKr
https://pbs.twimg.com/media/HQyfABta8AAq3x7.jpgView originalTencent AI ·
Introducing Hy4 preview ◆ Open weights: 770B MoE, 49B active, 1M context ◆ Built for real work: code, docs and analysis, scientific research ◆ Hy4 preview helped optimize its own training pipeline, kernels included ◆ We ship upgrades, not price hikes Free in @WorkBuddy_AI https://t.co/sCb2eYj6Vq
https://pbs.twimg.com/media/HQyn5vtaoAAGLLB.jpgView originalTencent Hy ·
We compressed Hy4-preview from 1.5TB to ～200GiB GGUF and it still works well ! Meet MIX-STQ1_0.The trick isn’t just going low, it’s deciding where: calibration data picks each layer’s bit-width, some down to 1.31-bit STQ1_0, some up to 2.06-bit IQ2_XXS. Same budget, lower https://t.co/xK8d3BH5br
https://pbs.twimg.com/media/HQ3cZjubEAAYzkS.jpghttps://pbs.twimg.com/media/HQ3cZjsa4AAgvNH.pngView originalReddit
Tencent/Hy4-preview 770B-A49B weight dropped
Beamsters ·
View originalTencent AI "Introducing Hy4 preview ◆ Open weights: 770B MoE, 49B active, 1M context ◆ Built for real work: code, docs and analysis, scientific research ◆ Hy4 preview helped optimize its own training pipeline, kernels included" ➡️ new model for you? :P
Koala_Confused ·
View originalTencent Hy4 Preview is live in Command Code.
maedahbatool ·
Tencent Hy4 Preview is live in Command Code 🐐 770B mode, 49B active, 1M context. `cmd update` available in GOAT and above plans. Our early internal benchmarks show strong perf yet cheaper per task compared to GLM 5.3 Flash. Y'all share how it goes.
View originalYouTube
Hy4 Preview: Tencent’s 780B Open Agent Model Is Here
RepoChad
View originalHy4 Preview: Built for Productivity and Beats GLM 5.3, DeepSeek, Opus
Fahd Mirza
View originalTencent Hy4 Preview: 770B Parameters, 49B Active, a 200-GiB GGUF, and Day-Zero vLLM
AI tech
View original