跳轉到主要內容

Hy3

Tencent開源權重Apache 2.0 · 商用許可

描述

Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and a 3.8B MTP layer, developed by the Tencent Hy Team. Following the Hy3 Preview launch in late April, the team scaled up post-training with higher-quality data and RL, gathering feedback from 50+ products. Hy3 outperforms similar-size models and rivals flagship open-source models with 2-5x the parameters, with strong gains in reasoning, agentic, and long-context tasks. It uses 80 layers (plus 1 MTP layer), 64 GQA attention heads (8 KV heads, head dim 128), a 4096 hidden size, 192 experts with top-8 activated, a 256K context window, and BF16 precision. Hy3 is a hybrid-thinking model supporting configurable reasoning effort (no_think, low, high), and emphasizes production-grade tool-call and output-format stability, reduced hallucination, and reliable multi-turn intent tracking.

發布日期
2026-07-06
參數規模
295.0B
上下文長度
262K
支援模態
text

能力雷達圖

2
general
60
coding
4
reasoning
77
science估算
70
agents
0
multimodal

缺少專門科學評測時,Science 由 LLM Stats 科學得分或推理能力估算。

排行榜排名

領域#排名分數來源
智慧體能力模型榜68
48.0
LS

基準測試分數 (LLM Stats)

(LLM Stats (zeroeval))

Agents

WildClawBench53.6%自報
Toolathlon48.5%自報

Chemistry

SuperChem54.9%自報

Code

Claw-Eval68.5%自報
SkillsBench55.3%自報
NL2Repo45.6%自報
DeepSWE28.0%自報
CL-bench23.8%自報
CL-bench (Life)17.0%自報

Math

USAMO 202630.24 / 42自報
IMO-AnswerBench90.0%自報
ArXivMath52.2%自報
MathArena Apex38.7%自報
HorizonMath7.1%自報

Physics

PHYBench77.4%自報
CMT-Benchmark37.9%自報

Reasoning

GPQANYU + Cohere + Anthropic (2023)90.4%自報
BrowseCompOpenAI (2025)84.2%自報
MCP Atlas79.1%自報
SWE-Bench Verified78.0%自報
SWE-bench Multilingual75.8%自報
FrontierScience Olympiad74.8%自報
AA-LCR73.4%自報
Terminal-Bench 2.171.7%自報
SWE-Bench ProPrinceton NLP (2024)57.9%自報
Humanity's Last Exam (with tools, text-only)53.2%自報
Humanity's Last Exam (no tools, text-only)47.0%自報
APEX-Agents25.6%自報
FrontierScience Research21.3%自報

Search

DeepSearchQA91.0%自報
WideSearch76.4%自報

AA 評測指數

(Artificial Analysis)

暫無 AA 評測資料

LLM Stats 分類評分

(LLM Stats (zeroeval))
Math
4
Reasoning
2
General
2
Biology
90
Physics
80
Search
80
Frontend Development
80
Long Context
70
Chemistry
70
Tool Calling
70
Science
60
Agents
60
Code
60
Knowledge
50
Coding
50

定價

輸入價格$0.136 / 1M tokens
輸出價格$0.555 / 1M tokens
混合價格(3:1)$0.25 / 1M tokens
快取讀取價格$0.033 / 1M tokens

速度

Tokens/秒80.8
首Token延遲2.83s
首回答延遲27.58s

供應商價格排行

供應商價格排行

18 個供應商

最便宜: NanoGPT最貴: AI/ML API
供應商輸入輸出
1NanoGPT最便宜
$0.066
$0.26
2OpenRouter
$0.132
$0.528
3Kilo Gateway
$0.132
$0.528
4DevPass (LLM Gateway)
$0.132
$0.528
5Eden AI
$0.132
$0.528
6LLM Gateway
$0.132
$0.528
7Tencent主要
$0.136
$0.555
8DeepInfra
$0.14
$0.58
9OpenCode Go
$0.14
$0.58
10Novita AI
$0.14
$0.58
11Requesty
$0.14
$0.58
12Vercel AI Gateway
$0.14
$0.58
13Jalapeno Cloud
$0.14
$0.58
14Kenari
$0.1543
$0.6286
15AIHubMix
$0.1562
$0.6248
16CrossModel
$0.16
$0.64
17OrcaRouter
$0.18
$0.59
18AI/ML API
$0.2116
$0.8464

比較該模型在不同 API 供應商之間的定價。

外部連結