关于Hy3-preview
Hy3 preview is a 295B-parameter Mixture-of-Experts (MoE) language model from Tencent Hunyuan, built for production-grade agent workloads. With only 21B parameters activated per token and native 256K context support, it handles complex tasks like cross-file code refactoring, long-document analysis, and multi-step tool use, rather than just generating fluent dialogue. Hy3 scores near state-of-the-art on SWE-bench Verified and advanced STEM benchmarks, while offering three inference modes (no_think, think_low, think_high) to dynamically trade off latency and reasoning depth. Its sparse activation architecture delivers competitive intelligence at a significantly lower token cost.
Leverage Hy3-preview’s 295B-parameter MoE architecture and 256K context for production-grade agentic workflows and deep reasoning.
Repository-Scale Refactoring
Execute complex architectural changes across massive codebases using native 256K context support.
Use Case Example:
"Migrated a distributed Go backend from REST to gRPC, updating service definitions and client libraries across 40+ repositories in a single pass."
Autonomous DevOps Agents
Power agents that navigate environments, use multi-step tools, and solve infrastructure issues autonomously.
Use Case Example:
"Deployed an agent to resolve a memory leak in a Rust-based embedded system by analyzing core dumps and applying a validated firmware patch."
PhD-Level STEM Reasoning
Tackle advanced scientific challenges in math and biology using high-depth reasoning modes.
Use Case Example:
"Formulated a formal proof for a fluid dynamics theorem using 'think_high' mode to validate complex boundary conditions for a research paper."
Intelligent Document Auditing
Analyze lengthy technical or legal documents to detect logical gaps and hidden risks with high precision.
Use Case Example:
"Scanned a 150-page semiconductor schematic and its manual to identify a power-sequencing logic error before the fabrication phase."
元数据
规格
州
Deprecated
建筑
Mixture-of-Experts
校准的
不
专家混合
是
总参数
80B
激活的参数
21B
推理
不
精度
FP8
上下文长度
262K
最大输出长度
与其他模型进行比较
看看这个模型与其他模型相比表现如何。
Tencent
chat
Hunyuan-MT-7B
上下文长度：
33K
最大输出长度：
33K
Input:
$
0.0
/ M Tokens
Output:
$
0.0
/ M Tokens
Tencent
chat
Hunyuan-A13B-Instruct
上下文长度：
131K
最大输出长度：
131K
Input:
$
0.14
/ M Tokens
Output:
$
0.57
/ M Tokens
Tencent
chat
Hy3
上下文长度：
262K
最大输出长度：
262K
Input:
$
0.132
/ M Tokens
Output:
$
0.528
/ M Tokens
Tencent
chat
Hy3-preview
上下文长度：
262K
最大输出长度：
Input:
$
0.066
/ M Tokens
Output:
$
0.26
/ M Tokens
