Hy3 Preview
Neural Network
Hy3 Preview is a Tencent Mixture-of-Experts model for agent tasks and production with adjustable reasoning levels
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 262 144 tokens
- Max answer length
- 235 929 tokens
- Reasoning effort
- none, low, high
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Hy3 Preview?
Hy3 Preview is a Tencent model based on the Mixture-of-Experts architecture: only a portion of experts are activated, making computations efficient, and the model is designed for industrial use. Its key feature is switchable reasoning depth: disabled, low, and high modes provide quick answers when sufficient, and detailed analysis for complex multi-step tasks. The model is specifically tailored for agent scenarios where a chain of actions is more important than a single response. In BotHub, it is accessible without a VPN or foreign card: you pay with a Russian card in rubles only for consumed tokens, with no subscription. Over 250 models are available in the same window, and a unified OpenAI-compatible API allows you to switch models in a multi-model system without rewriting integration. For companies, we offer contracts, invoices, and an admin panel with limits. It is useful when building agents, automating workflows, handling production loads, solving complex tasks in high mode, saving on routine tasks with reasoning disabled, and comparing responses with other models before choosing.Frequently asked questions about Hy3 Preview
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
hy3-preview is a Tencent text model that can reason before answering and accept images and documents as input. It is suitable for coding tasks: writing, refactoring, and debugging, analyzing long files, STEM tasks, and logical chains, as well as for agents with function calling.
The context window is 262,144 tokens, with a maximum output of 235,929 tokens. This is enough for large technical text, detailed code base analysis, or long translations in full without splitting. You control the response length via a request parameter or prompt.
Yes, reasoning before answering is confirmed in our data. The model first builds a chain of steps and then provides the result — this significantly helps with logic, math, and code debugging tasks where the path to the result is as important as the result itself.
Yes, both. The model calls functions and returns structured JSON, which is easy to parse on the service side. In BotHub, this works via a unified OpenAI-compatible API: connect once and switch between models without rewriting integration.
Images and documents — yes: you can send an interface screenshot, diagram, or file and ask to analyze the content. There is no mention of audio or video support in our data, so for such formats, it is more convenient to choose a specialized model in BotHub in the same window.
We do not have data on the quality of performance for specific languages, so we won't make promises — test it on your task. In BotHub, it's easy: access from Russia without a VPN or foreign card, pay in rubles for tokens used, and other models are available nearby for comparison.