Kimi K2.5 vs DeepSeek V3.2
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | Kimi K2.5 | DeepSeek V3.2 |
|---|---|---|
| Provider |  moonshotai |  DeepSeek |
| pontuação média | 52.0 | 53.0 |
| Preço de entrada | $0.38 | $0.26 |
| Preço de saída | $1.72 | $0.38 |
| Janela de contexto | 262K tokens (~131 books) | 164K tokens (~82 books) |
| Lançado em | 2026-01-27 | 2025-12-01 |
| Código aberto | Open Source | Open Source |
Pontuações de benchmark
18 benchmarks · Kimi K2.5: 16, DeepSeek V3.2: 2
| Benchmark | Categoria | Kimi K2.5 | DeepSeek V3.2 |
|---|---|---|---|
| ARC-AGI | reasoning | 65.3 | 57.0 |
| ARC-AGI-2 | reasoning | 11.8 | 4.0 |
| Artificial Analysis — Agentic Index | speed | 58.9 | 52.9 |
| Artificial Analysis — Coding Index | speed | 39.5 | 36.7 |
| Artificial Analysis — Quality Index | speed | 46.8 | 41.7 |
| Chess Puzzles | knowledge | 12.0 | 14.0 |
| FrontierMath-2025-02-28-Private | math | 27.9 | 22.1 |
| FrontierMath-Tier-4-2025-07-01-Private | math | 4.2 | 2.1 |
| GPQA diamond | knowledge | 83.5 | 77.9 |
| OpenCompass — AIME2025 | math | 91.9 | 93.0 |
| OpenCompass — GPQA-Diamond | knowledge | 88.1 | 84.6 |
| OpenCompass — HLE | knowledge | 28.6 | 23.2 |
| OpenCompass — IFEval | language | 93.9 | 89.7 |
| OpenCompass — LiveCodeBenchV6 | coding | 80.6 | 75.4 |
| OpenCompass — MMLU-Pro | knowledge | 86.2 | 85.8 |
| OTIS Mock AIME 2024-2025 | math | 92.2 | 87.8 |
| SimpleQA Verified | knowledge | 33.9 | 27.5 |
| Terminal Bench | coding | 43.2 | 39.6 |