Kimi K2.5 vs Kimi K2 Thinking
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | Kimi K2.5 | Kimi K2 Thinking |
|---|---|---|
| Provider |  moonshotai | moonshotai |
| pontuação média | 52.0 | 53.3 |
| Preço de entrada | $0.38 | $0.60 |
| Preço de saída | $1.72 | $2.50 |
| Janela de contexto | 262K tokens (~131 books) | 262K tokens (~131 books) |
| Lançado em | 2026-01-27 | 2025-11-06 |
| Código aberto | Open Source | Open Source |
Pontuações de benchmark
16 benchmarks · Kimi K2.5: 14, Kimi K2 Thinking: 2
| Benchmark | Categoria | Kimi K2.5 | Kimi K2 Thinking |
|---|---|---|---|
| APEX-Agents | agentic | 14.4 | 4.0 |
| Chess Puzzles | knowledge | 12.0 | 20.0 |
| FrontierMath-2025-02-28-Private | math | 27.9 | 21.4 |
| FrontierMath-Tier-4-2025-07-01-Private | math | 4.2 | 0.1 |
| GPQA diamond | knowledge | 83.5 | 79.0 |
| OpenCompass — AIME2025 | math | 91.9 | 94.1 |
| OpenCompass — GPQA-Diamond | knowledge | 88.1 | 82.7 |
| OpenCompass — HLE | knowledge | 28.6 | 21.3 |
| OpenCompass — IFEval | language | 93.9 | 92.4 |
| OpenCompass — LiveCodeBenchV6 | coding | 80.6 | 77.1 |
| OpenCompass — MMLU-Pro | knowledge | 86.2 | 84.3 |
| OTIS Mock AIME 2024-2025 | math | 92.2 | 83.0 |
| PostTrainBench | knowledge | 10.3 | 7.3 |
| SimpleQA Verified | knowledge | 33.9 | 31.6 |
| Terminal Bench | coding | 43.2 | 35.7 |
| WeirdML | coding | 45.6 | 42.8 |