Kimi K3vsGPT-5.6 Sol
Kimi K3 与 GPT-5.6 Sol 在 30 个共同 benchmark 中整体接近：Kimi K3 领先 15 项，GPT-5.6 Sol 领先 15 项，持平 0 项，平均分差 -2.45。
Kimi K3
Moonshot AI · 2026-07-16 · 推理大模型
GPT-5.6 Sol
OpenAI · 2026-06-26 · 推理大模型
Kimi K315 项(50%)(50%)15 项GPT-5.6 Sol
评测分数
按能力类目分组，每组内按分差大小排列；共 30 项。
法律
Kimi K3 领先 2/3| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Harvey's Legal Agent Benchmark | 10.8310 / 43Max (With Tools) | 2.5029 / 43Max (With Tools) | +8.33 |
| Harvey Lab-AA | 94.642 / 44Max (With Tools) | 87.1817 / 44Max (With Tools) | +7.46 |
| Legal Research Bench | 44.2011 / 42Max (With Tools) | 48.087 / 42Max (With Tools) | -3.88 |
跨应用与工具编排
Kimi K3 领先 2/3| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| APEX-Agents | 415 / 7Max (With Tools) | 56.703 / 7Max (With Tools) | -15.70 |
| MCPMark-Verified | 94.501 / 8Max (With Tools) | 92.912 / 8Max (With Tools) | +1.59 |
| MCP-Atlas | 82.309 / 44Max (With Tools) | 81.8013 / 44Max (With Tools) | +0.50 |
金融
Kimi K3 领先 2/3| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| EMB (Excel Modeling) | 66.4015 / 42Max (With Tools) | 72.345 / 42Max (With Tools) | -5.94 |
| Tax Agent Bench | 68.677 / 20Max (With Tools) | 67.958 / 20Max (With Tools) | +0.72 |
| Finance Agent v2 | 54.4018 / 43Max (With Tools) | 53.7621 / 43Max (With Tools) | +0.64 |
临床工作与医疗决策
Kimi K3 领先 2/2| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| MedCode | 48.8819 / 64Max (With Tools) | 43.9732 / 64Max (With Tools) | +4.91 |
| MedScribe | 87.9610 / 66Max (With Tools) | 85.2320 / 66Max (With Tools) | +2.73 |
图像感知与视觉推理
GPT-5.6 Sol 领先 2/2| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| ZeroBench Main | 414 / 6Max (With Tools) | 531 / 6Max (With Tools) | -12 |
| BabyVision | 85.705 / 9Max (With Tools) | 88.904 / 9Max (With Tools) | -3.20 |
客服与业务流程
Kimi K3 领先 2/2| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Public Benefits Bench v1.1 | 68.277 / 26Max (With Tools) | 66.5110 / 26Max (With Tools) | +1.76 |
| SAGE | 54.263 / 64Max (With Tools) | 52.564 / 64Max (With Tools) | +1.70 |
数学
胶着 2/2| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| FrontierMath v2 | 72.1813 / 58Max (No Tools) | 89.121 / 58Max (No Tools) | -16.94 |
| ProofBench v1.1 (Lean 4) | 876 / 28Max (With Tools) | 837 / 28Max (With Tools) | +4 |
程序生成与编辑
胶着 2/2| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Program Bench | 17.5012 / 15Max (With Tools) | 238 / 15Max (With Tools) | -5.50 |
| Vibe Code Bench v1.1 | 84.9610 / 61Max (With Tools) | 80.5017 / 61Max (With Tools) | +4.46 |
专业数据分析
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| AA-AnalystAgent | 38.7514 / 29Max (With Tools + Internet) | 47.507 / 29Max (With Tools + Internet) | -8.75 |
仓库修复与多文件工程
Kimi K3 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| NL2Repo-Bench | 586 / 16Max (With Tools) | 56.809 / 16Max (With Tools) | +1.20 |
写作与创意表达
Kimi K3 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Creative Writing | 2,0825 / 110Normal (No Tools) | 1,9729 / 110Normal (No Tools) | +110.70 |
办公与文档流程
Kimi K3 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| AutomationBench | 46.7010 / 25Max (With Tools) | 45.8011 / 25Max (With Tools) | +0.90 |
多轮记忆与持续上下文
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Context Arena | 71.7559 / 126Max (No Tools) | 97.632 / 126Max (No Tools) | -25.88 |
文档与图表理解
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Chartography | 68.108 / 11Max (With Tools) | 79.904 / 11Max (With Tools) | -11.80 |
漏洞发现与分析
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| CyberGym | 808 / 11Max (With Tools) | 84.505 / 11Max (With Tools) | -4.50 |
科学计算与科研编程
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Terminal-Bench-Science 0.1 | 7.1014 / 18Max (With Tools) | 22.406 / 18Max (With Tools) | -15.30 |
算法与竞赛编程
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| IOI (Vals v2) | 48.9418 / 26Max (With Tools) | 91.173 / 26Max (With Tools) | -42.23 |
维护、优化与工程管理
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Code Migration | 16.1034 / 43Max (With Tools) | 52.927 / 43Max (With Tools) | -36.82 |
自主开发与终端任务
GPT-5.6 Sol 领先 1/1| 评测项 | Kimi K3 | GPT-5.6 Sol | 分差 |
|---|---|---|---|
| Terminal-Bench 3.0 | 17.708 / 11Max (With Tools) | 34.402 / 11Max (With Tools) | -16.70 |
规格对比
| 字段 | Kimi K3 | GPT-5.6 Sol |
|---|---|---|
| 发布机构 | Moonshot AI | OpenAI |
| 发布时间 | 2026-07-16 | 2026-06-26 |
| 模型类型 | 推理大模型 | 推理大模型 |
| 架构 | MoE 架构 | 稠密模型 |
| 参数规模 | 2.8万亿 | 暂无数据 |
| 上下文长度 | 1M | 1.05M |
| 最大输出 | 1M | 128K |
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 | GPT-5.6 Sol |
|---|---|---|
| 文本输入 | ¥20 / 1M tokens | $4 / 1M tokens |
| 文本输出 | ¥100 / 1M tokens | $20 / 1M tokens |
| 缓存读取 | ¥2 / 1M tokens | $0.4 / 1M tokens |
| 缓存写入 | 暂无公开价格 | $5 / 1M tokens |
小结
- Kimi K3在以下类目领先:法律 (2/3)、跨应用与工具编排 (2/3)、金融 (2/3)、临床工作与医疗决策 (2/2)、客服与业务流程 (2/2)、仓库修复与多文件工程 (1/1)、写作与创意表达 (1/1)、办公与文档流程 (1/1)
- GPT-5.6 Sol在以下类目领先:图像感知与视觉推理 (2/2)、专业数据分析 (1/1)、多轮记忆与持续上下文 (1/1)、文档与图表理解 (1/1)、漏洞发现与分析 (1/1)、科学计算与科研编程 (1/1)、算法与竞赛编程 (1/1)、维护、优化与工程管理 (1/1)、自主开发与终端任务 (1/1)
- 胶着类目:数学、程序生成与编辑
30 个共同 benchmark 上，GPT-5.6 Sol 平均高出 2.45 分。
单项差距最大的 benchmark：Creative Writing — Kimi K3 2,082，GPT-5.6 Sol 1,972（分差 +110.70）。
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。