See key specs and per-benchmark scores for each model/mode. Scroll horizontally for all columns. 当前对比 2 个模型的评测数据与核心参数。
Kimi K3
Moonshot AI
Each axis is a category average, normalized to a 100-point radar.
Relative edge: 编程与软件工程 +8.6 / Relative gap: 数学推理 -4.9
Relative edge: 数学推理 +4.9 / Relative gap: 编程与软件工程 -8.6
Method: for each model and benchmark, the chart first averages all scores in the current mode scope instead of taking the best score, then averages those benchmark scores within each category. Only benchmarks with at least two selected models scored are included; missing values are not counted as zero.
Best overall
Kimi K3 · 61.96
Best single
Kimi K3 · GPQA Diamond 93.50
Modality coverage
Kimi K3 · 3 modalities
Head to head
11
Benchmarks
8
Wins
3
Losses
+2.47
Average diff
Compare benchmark results across thinking modes and tool usage.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Complete scores for each model/mode across selected benchmarks.
11 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.
| Benchmark | Kimi K3 | Qwen3.8-Max |
|---|---|---|
HLE 综合评估 | 56.00Thinking Level · High ｜ Tools | 56.20Thinking Level · Extra High ｜ Tools |
GPQA Diamond 科学与综合推理 | 93.50Thinking Level · High | 92.60Thinking Level · Extra High |
AutomationBench AI Agent - 工具使用 | 30.80Thinking Level · High ｜ Tools | 27.30Thinking Level · Extra High ｜ Tools |
Terminal-Bench 2.1 AI Agent - 工具使用 | 88.30Thinking Level · High ｜ Tools | 86.60Thinking Level · Extra High ｜ Tools |
Toolathlon-Verified AI Agent - 工具使用 | 76.50Thinking Level · High ｜ Tools | 72.50Thinking Level · Extra High ｜ Tools |
DeepSWE 编程与软件工程 | 67.50Thinking Level · High ｜ Tools | 56.60Thinking Level · Extra High ｜ Tools |
FrontierSWE 编程与软件工程 | 81.20Thinking Level · High ｜ Tools | 73.50Thinking Level · Extra High ｜ Tools |
MLS Bench 编程与软件工程 | 48.30Thinking Level · High ｜ Tools | 41.00Thinking Level · Extra High ｜ Tools |
Agents' Last Exam Agent能力评测 | 28.30Thinking Level · High ｜ Tools | 27.00Thinking Level · Extra High ｜ Tools |
39.02Thinking Level · High | 46.34Thinking Level · Extra High | |
FrontierMath v2 数学推理 | 72.18Thinking Level · High | 74.74Thinking Level · Extra High |
Side-by-side input/output token pricing
Licensing, MoE architecture, and multi-modality support.
| Features & specs | Kimi K3Moonshot AI | Qwen3.8-Max阿里巴巴 |
|---|---|---|
Core specsRelease | 2026-07-16 | 2026-08-03 |
Context length | 1M | 1M |
Parameters | 2.8T | 2.4T |
Active parameters | 104B | 95B |
Max output | 1048576 | 131072 |
MoE | Yes | Yes |
LicenseCode Open Source | Open Source · Kimi K3 License | Open Source · Qwen3.8-Max License |
Weights Open Source | Open Source · Kimi K3 License | Open Source · Qwen3.8-Max License |
Commercial use | 有条件免费商用授权 | 有条件免费商用授权 |
Local deploymentWeight size | 约 1.42 TiB（MXFP4，96 个 Safetensors 分片） | 约 4.89 TB（BF16，213 个 safetensors 分片） |
VRAM for weights | ≈ 1454 GB (weights only, excludes KV cache) | ≈ 4890 GB (weights only, excludes KV cache) |
Weights | Hugging Face | Hugging Face |
Source repo | GitHub | Not provided |
Modality supportText Input/Output | / | / |
Image Input/Output | / | / |
Video Input/Output | / | / |
ResourcesPaper / report | Kimi K3 技术报告 | 阿里Qwen3.8-Max模型发布：2.4万亿参数，瞄准复杂长程任务 |
Qwen3.8-Max
阿里巴巴
