Hy4 previewvsKimi K3
Hy4 preview 与 Kimi K3 在 2 个同模式 benchmark 中整体接近：Hy4 preview 领先 1 项，Kimi K3 领先 1 项，持平 0 项，平均分差 -0.19。另有 15 项测试模式不同，仅供参考。
Hy4 preview
腾讯AI实验室 · 2026-08-28 · 推理大模型
Kimi K3
Moonshot AI · 2026-07-16 · 推理大模型
Hy4 preview1 项(50%)(50%)1 项Kimi K3
评测分数
按能力类目分组，每组内按分差大小排列；共 2 项。
科学知识与推理
Hy4 preview 领先 1/1| 评测项 | Hy4 preview | Kimi K3 | 分差 |
|---|---|---|---|
| GPQA Diamond | 92.3023 / 253Thinking High (No Tools) | 91.9224 / 253Thinking High (No Tools) | +0.38 |
跨能力聚合指数
Kimi K3 领先 1/1| 评测项 | Hy4 preview | Kimi K3 | 分差 |
|---|---|---|---|
| Vals Index v2.1 | 49.9426 / 37Reported best (effort unspecified) | 50.7025 / 37Reported best (effort unspecified) | -0.76 |
测试模式不同的成绩
共 15 项。两款模型的公开成绩来自不同测试模式，分数不直接可比，不计入胜负和平均分差。
| 评测项 | Hy4 preview | Kimi K3 |
|---|---|---|
| BioMysteryBench | 71.30Thinking High (With Tools) | 72.59Max (With Tools) |
| Program Bench | 17.50Thinking High (With Tools) | 17.50Max (With Tools) |
| HLE | 55.40Thinking High (With Tools) | 59.80Max (With Tools) |
| PostTrain Bench | 35.60Thinking High (With Tools) | 36.60Max (With Tools) |
| AutomationBench | 32.10Thinking High (With Tools) | 46.70Max (With Tools) |
| Office QA Pro | 66.20Thinking High (With Tools) | 63.30Max (With Tools) |
| DeepSWE | 64.30Thinking High (With Tools) | 68.51Max (With Tools) |
| NL2Repo-Bench | 58.90Thinking High (With Tools) | 58Max (With Tools) |
| SWE-Marathon | 31.90Thinking High (With Tools) | 42Max (With Tools) |
| CritPt | 16.90Thinking High (No Tools) | 23.40Max (No Tools) |
| APEX-Agents | 37.10Thinking High (With Tools) | 41Max (With Tools) |
| Job Bench | 61.70Thinking High (With Tools) | 54.30Max (With Tools) |
| MCP-Atlas | 83.70Thinking High (With Tools) | 82.30Max (With Tools) |
| Toolathlon-Verified | 74.10Thinking High (With Tools) | 76.50Max (With Tools) |
| CyberGym | 78.40Thinking High (With Tools) | 80Max (With Tools) |
规格对比
| 字段 | Hy4 preview | Kimi K3 |
|---|---|---|
| 发布机构 | 腾讯AI实验室 | Moonshot AI |
| 发布时间 | 2026-08-28 | 2026-07-16 |
| 模型类型 | 推理大模型 | 推理大模型 |
| 架构 | MoE 架构 | MoE 架构 |
| 参数规模 | 7700亿 | 2.8万亿 |
| 上下文长度 | 1M | 1M |
| 最大输出 | 暂无数据 | 1M |
API 调用价格
价格优先使用 DataLearner 配置的 API 记录；缺失项不做推测。
| 价格项 | Hy4 preview | Kimi K3 |
|---|---|---|
| 文本输入 | ¥6 / 1M tokens | ¥20 / 1M tokens |
| 文本输出 | ¥18 / 1M tokens | ¥100 / 1M tokens |
| 缓存读取 | ¥0.3 / 1M tokens | ¥2 / 1M tokens |
小结
- Hy4 preview在以下类目领先:科学知识与推理 (1/1)
- Kimi K3在以下类目领先:跨能力聚合指数 (1/1)
2 个同模式 benchmark 上，Kimi K3 平均高出 0.19 分。
单项差距最大的 benchmark：Vals Index v2.1 — Hy4 preview 49.94，Kimi K3 50.70（分差 -0.76）。
本页正文由结构化模型、价格与 benchmark 数据生成，不使用实时 LLM 撰写。