Kimi K3vsKimi K2.6
在 1 个同模式 benchmark 中，Kimi K3 整体领先：Kimi K3 领先 1 项，Kimi K2.6 领先 0 项，持平 0 项，平均分差 +357.80。另有 20 项测试模式不同，仅供参考。
Kimi K31 项(100%)(0%)0 项Kimi K2.6
评测分数
按能力类目分组，每组内按分差大小排列；共 1 项。
写作与创意表达
Kimi K3 领先 1/1| 评测项 | Kimi K3 | Kimi K2.6 | 分差 |
|---|---|---|---|
| Creative Writing | 2,0825 / 112Normal (No Tools) | 1,72527 / 112Normal (No Tools) | +357.80 |
测试模式不同的成绩
共 20 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 | Kimi K3 | Kimi K2.6 |
|---|---|---|
| Kimi Code Bench 2.0 | 72.90Max (With Tools) | 50.90Thinking (With Tools) |
| Program Bench | 17.50Max (With Tools) | 48.30Thinking (With Tools) |
| LiveBench | 79.19Reported best (effort unspecified) | 70.54Thinking (No Tools) |
| OSWorld-Verified | 84.80Max (With Tools) | 73.10Thinking (With Tools) |
| GDP.pdf | 22Max (No Tools) | 13Thinking (No Tools) |
| BrowseComp | 91.20Max (With Tools + Internet) | 83.20Thinking (With Tools + Internet) |
| HLE | 59.80Max (With Tools) | 54Thinking (With Tools + Internet) |
| Harvey Lab-AA | 94.64Max (With Tools) | 84.12Thinking (With Tools) |
| FrontierMath Tier 4 v2 | 39.02Max (With Tools) | 25.64Reported best (effort unspecified) |
| FrontierMath v2 | 72.18Max (With Tools) | 57.19Reported best (effort unspecified) |
| Context Arena | 71.75Max (No Tools) | 51.88Normal (No Tools) |
| MLS Bench | 48.30Max (With Tools) | 26.70Thinking (With Tools) |
| SciCode | 59.50Max (No Tools) | 51.50Thinking (No Tools) |
| CritPt | 23.40Max (No Tools) | 1.40Normal (No Tools) |
| GPQA Diamond | 91.92Thinking High (No Tools) | 90.50Thinking (No Tools) |
| τ³-Banking | 41.60Thinking Low (With Tools) | 23.30Thinking (With Tools) |
| MCP-Atlas | 82.30Max (With Tools) | 69.40Thinking (With Tools) |
| MCPMark-Verified | 94.50Max (With Tools) | 72.80Thinking (With Tools) |
| Toolathlon-Verified | 76.50Max (With Tools) | 58Thinking (With Tools) |
| MMMU-Pro | 83.40Max (With Tools) | 79.40Thinking (No Tools) |
规格对比
| 字段 | Kimi K3 | Kimi K2.6 |
|---|---|---|
| 发布机构 | Moonshot AI | Moonshot AI |
| 发布时间 | 2026-07-16 | 2026-04-20 |
| 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 | 2.8万亿 | 1万亿 |
| 上下文长度 | 1M | 256K |
| 最大输出 | 1M | 暂无数据 |
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 | Kimi K3 | Kimi K2.6 |
|---|---|---|
| 文本输入 | ¥20 / 1M tokens | $0.95 / 1M tokens |
| 文本输出 | ¥100 / 1M tokens | $4 / 1M tokens |
| 缓存读取 | ¥2 / 1M tokens | $0.16 / 1M tokens |
小结
- Kimi K3在以下类目领先:写作与创意表达 (1/1)
1 个同模式 benchmark 上，Kimi K3 平均高出 357.80 分。
单项差距最大的 benchmark：Creative Writing — Kimi K3 2,082，Kimi K2.6 1,725（分差 +357.80）。
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。