Hy3
Neural Network
Hy3 is a Mixture-of-Experts model by Tencent for reasoning and agentic scenarios; Hy3 API is available in BotHub.
Max answer length
(in tokens)
Context size
(in tokens)
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Hy3?
Hy3 is a large language model by Tencent based on the Mixture-of-Experts architecture: 295B parameters, with about 21B activated per request, routing across 192 experts with a top-8 selection. This design is focused on reasoning, agentic scenarios, and real-world production workloads, while adjustable reasoning effort allows choosing between faster or more thorough responses. In BotHub, the model is accessible without a VPN or foreign card: you pay in rubles per token, no subscription required, and unused tokens do not expire. Over 250 models are available in the same window for easy comparison, and a unified OpenAI-compatible API lets you switch models without rewriting integrations. Traffic is encrypted via AES-GCM, and companies get access to contracts, invoices, electronic document management, and an admin panel with limits. Analysts can use Hy3 for multi-step reasoning, developers as an agent core for executing chains of steps, product teams for automating routine tasks and processing requests, and researchers to run a single task through multiple models to select the best one.Frequently asked questions about Hy3
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 is a Tencent text model with a 262,144 token context window. It analyzes long documents, writes and refactors code, helps with debugging, solves reasoning and STEM tasks, and calls external functions. Suitable for analytics, development support, and working with large text datasets.
In a single response, Hy3 generates up to 128,000 tokens—enough for a voluminous document, detailed analysis, or a large code module in its entirety without splitting. If the response cuts off, simply ask it to continue: the 262,144 token dialogue context is preserved.
Yes, Hy3 has a reasoning mode: before the final answer, the model builds a chain of steps, checks intermediate conclusions, and only then formulates the result. This significantly helps with math, logic puzzles, code analysis, and situations where an argumentative conclusion is more important than a guess.
Yes, Hy3 supports function calling and structured JSON output. The model decides when to call your tool and returns arguments in the specified schema. Through the unified OpenAI-compatible BotHub API, this connects to agents and services without rewriting integrations.
Images and documents—yes: Hy3 analyzes screenshots, diagrams, charts, and files, answering based on their content. We have no data on audio and video, so expect support for images and text attachments. For audio or video, it is convenient to switch to a specialized model in BotHub.
We do not have data on the quality of Hy3's performance with specific languages, so we cannot make any promises. The easiest way to check is with your own task: you pay per token, they do not expire, and you can compare the response with another model in the same window.