Kimi K2.5
| Released | 2026-01-27 |
|---|---|
| Price per 1M | $0.60 in · $3.00 out |
| Overall score | 53.2 · rank 89 |
| Context window | 262,144 tokens |
| Provider | Moonshot AI |
| Weights | Open weights |
Scores
| Dimension | Score | Rank | Benchmarks |
|---|---|---|---|
| Overall: Balanced | 53.2 | 89 | 31 |
| Overall: Coding | 52.5 | 87 | 16 |
| Overall: Reasoning | 53.1 | 89 | 16 |
| Overall: Agentic work | 53.1 | 91 | 14 |
| Overall: Frontend | 48.8 | 101 | 7 |
| Coding | 54.4 | 81 | 3 |
| Agents | 52.5 | 90 | 3 |
| Reasoning | 53.4 | 91 | 8 |
| Math | 53.5 | 81 | 4 |
| Knowledge | 51.9 | 83 | 4 |
| Frontend | 44.7 | 95 | 2 |
| General | 55.3 | 65 | 5 |
| Multimodal | 56.3 | 40 | 2 |
| Value vs price | +4.9 | 75 |
Configurations
| Configuration | Balanced | Benchmarks |
|---|---|---|
| Unspecified setting | 53.6 ±1.5 | 19 |
| Non-reasoning | 50.8 ±5.6 | 3 |
| Thinking | 52.7 ±1.4 | 16 |
Evidence
Coding 54.4 ±2.9
| Benchmark | Result | Source | Date |
|---|---|---|---|
| SciCode | 49.0% | Epoch AI | 2026-11-11 |
| LiveCodeBench (Vals run) | 83.9% | Vals AI | 2026-09-01 |
| SWE-bench Verified (Epoch run) | 73.8% | Epoch AI | 2026-02-17 |
| SWE-bench Verified (Vals run) | 70.0% | Vals AI | 2026-09-01 |
Agents 52.5 ±4.0
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Vending-Bench 2 | $1,198 | Epoch AI | 2026-11-11 |
| OSWorld-Verified | 63.3% | OSWorld-Verified | 2026-01-31 |
| Terminal-Bench 2.0 | 43.2% | Epoch AI | 2026-01-27 |
| Terminal-Bench 2.1 (Vals run) | 41.9% | Vals AI | 2026-10-01 |
| OSWorld (v1) | 63.3% | Epoch AI | 2026-11-11 |
Reasoning 53.4 ±1.6
| Benchmark | Result | Source | Date |
|---|---|---|---|
| ARC-AGI-2 | 11.8% | Epoch AI | 2026-11-11 |
| Extended NYT Connections | 69.9% | Lech Mazur | 2026-11-10 |
| ARC-AGI-2 (ARC Prize verified) | 11.8% | ARC Prize | 2026-11-11 |
| CritPt | 3.1% | Epoch AI | 2026-11-11 |
| ARC-AGI-1 (ARC Prize verified) | 65.3% | ARC Prize | 2026-11-11 |
| ARC-AGI-1 | 65.3% | Epoch AI | 2026-11-11 |
| EnigmaEval | 3.4% | Epoch AI | 2026-11-11 |
| GPQA Diamond | 87.6% | Epoch AI | 2026-02-02 |
| GPQA Diamond (Vals run) | 84.1% | Vals AI | 2026-09-01 |
| Kagi Reasoning Benchmark | 78.5% | Kagi | 2026-11-11 |
| SimpleBench | 46.8% | SimpleBench | 2026-02-17 |
| SimpleBench | 46.8% | Epoch AI | 2026-11-11 |
Math 53.5 ±1.8
| Benchmark | Result | Source | Date |
|---|---|---|---|
| FrontierMath Tiers 1-3 (2025 set) | 27.9% | Epoch AI | 2026-02-03 |
| FrontierMath Tier 4 (2025 set) | 4.2% | Epoch AI | 2026-02-02 |
| OTIS Mock AIME 2024-2025 | 92.2% | Epoch AI | 2026-02-02 |
| MathArena Final-Answer Competitions | 62.3% | MathArena | 2026-11-11 |
Knowledge 51.9 ±2.1
| Benchmark | Result | Source | Date |
|---|---|---|---|
| SimpleQA Verified | 34.3% | Epoch AI | 2026-08-27 |
| Humanity's Last Exam | 24.4% | Epoch AI | 2026-11-11 |
| MMLU-Pro (Vals run) | 85.9% | Vals AI | 2026-09-01 |
| Legal Research Bench | 15.9% | Vals AI | 2026-11-09 |
Frontend 44.7 ±0.7
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Vibe Code Bench v1.1 | 17.5% | Vals AI | 2026-11-09 |
| LMArena WebDev | 1,436 | LMArena | 2026-11-09 |
General 55.3 ±1.8
| Benchmark | Result | Source | Date |
|---|---|---|---|
| LLM Debate Benchmark | 1,445 | Lech Mazur | 2026-10-26 |
| EQ-Bench Creative Writing v3 | 1,579 | EQ-Bench | 2026-11-11 |
| LMArena Text | 1,450 | LMArena | 2026-11-09 |
| Creative Story-Writing (Lech Mazur) | 33.0% | Lech Mazur | 2026-10-23 |
| EQ-Bench Longform Creative Writing | 74.9% | EQ-Bench | 2026-11-11 |
Multimodal 56.3 ±2.3
| Benchmark | Result | Source | Date |
|---|---|---|---|
| MMMU-Pro (Vals run) | 84.3% | Vals AI | 2026-09-01 |
| LMArena Vision | 1,252 | LMArena | 2026-11-09 |
Results not used in the score
| Benchmark | Result | Source | Why not used |
|---|---|---|---|
| Kagi Reasoning Benchmark | 63.8% | Kagi | A stronger configuration result was used |
| LMArena Text | 1,431 | LMArena | A stronger configuration result was used |
| LMArena Vision | 1,236 | LMArena | A stronger configuration result was used |
| LMArena WebDev | 1,406 | LMArena | A stronger configuration result was used |
Names matched to this model
kimi-k2-5, kimi-k2p5