Kimi K2.5vsKimi K2
Across 9 shared benchmarks, Kimi K2.5 leads overall: Kimi K2.5 wins 8, Kimi K2 wins 1, with 0 ties and an average score difference of +25.61.
Kimi K2.58 wins(89%)(11%)1 winKimi K2
Benchmark scores
Grouped by capability, sorted by largest gap within each. 9 shared benchmarks.
General Knowledge
Kimi K2.5 3/4| Benchmark | Kimi K2.5 | Kimi K2 | Diff |
|---|---|---|---|
| ARC-AGI | 65.3031 / 65Thinking (No Tools) | 13.3057 / 65 | +52 |
| HLE | 50.2017 / 149Thinking (With Tools) | 4.70146 / 149 | +45.50 |
| GPQA Diamond | 87.6031 / 175Thinking (No Tools) | 75.1090 / 175 | +12.50 |
| MMLU Pro | 78.5064 / 124Thinking (No Tools) | 81.1051 / 124 | -2.60 |
Math and Reasoning
Kimi K2.5 3/3| Benchmark | Kimi K2.5 | Kimi K2 | Diff |
|---|---|---|---|
| AIME2025 | 96.1021 / 106Thinking (No Tools) | 5485 / 106 | +42.10 |
| Simple Bench | 46.8013 / 27Thinking (No Tools) | 26.3024 / 27 | +20.50 |
| FrontierMath - Tier 4 | 4.2040 / 80Normal (No Tools) | 0.0171 / 80 | +4.19 |
Coding and Software Engineer
Kimi K2.5 2/2| Benchmark | Kimi K2.5 | Kimi K2 | Diff |
|---|---|---|---|
| LiveCodeBench | 8514 / 118Thinking (No Tools) | 53.7084 / 118 | +31.30 |
| SWE-bench Verified | 76.8022 / 103Thinking (With Tools) | 51.8083 / 103 | +25 |
Specs
| Field | Kimi K2.5 | Kimi K2 |
|---|---|---|
| Publisher | Moonshot AI | Moonshot AI |
| Release date | 2026-01-27 | 2025-07-11 |
| Model type | Multimodal model | AI model |
| Architecture | MoE | MoE |
| Parameters | 10000.0 | 10000.0 |
| Context length | 256K | 131K |
| Max output | 16384 | 134144 |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Kimi K2.5 | Kimi K2 |
|---|---|---|
| Text input | 0.6 美元/100 万tokens | 0.6 美元/100 万tokens |
| Text output | 3 美元/100 万tokens | 2.5 美元/100 万tokens |
| Cache read | 0.1 美元/100 万tokens | Not public |
Summary
- Kimi K2.5leads in:General Knowledge (3/4), Math and Reasoning (3/3), Coding and Software Engineer (2/2)
On average across the 9 shared benchmarks, Kimi K2.5 scores 25.61 higher.
Largest single-benchmark gap: ARC-AGI — Kimi K2.5 65.30 vs Kimi K2 13.30 (+52).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.