Moonshot AIproprietary
Released: 2026-07-15MoonshotAI: Kimi K3
Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic workflows, and is particularly strong at...
Intelligence Index
91.5/ 100
Calibrated multi-domain compositeThroughput Speed
55tok/sec
Output streaming throughputTime To First Token
420ms
Initial chunk server latencyLiveBench Contam-Free
80.5%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning84.2%
Coding & Repo Repair82.4%
Mathematics (AIME)81.4%
Data Analysis & Tables77.8%
Language & Comprehension79.6%
Instruction Following80.5%
Sourced Benchmark Evaluations (60)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 57.7% | 93 | 2026-08-19 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 69.3% | 75 | 2026-08-12 | independent | Source ↗ |
| FrontierMath | Active | 23.8% | 98 | 2026-08-09 | independent | Source ↗ |
| PutnamBench | Active | 44.3% | 90 | 2026-08-08 | independent | Source ↗ |
| Omni-MATH | Deprecated | 63% | 10 | 2026-08-06 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 59.1% | 20 | 2026-08-06 | independent | Source ↗ |
| FrontierCode | Active | 50.4% | 98 | 2026-08-04 | independent | Source ↗ |
| Soohak | Active | 64.7% | 98 | 2026-08-04 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 68.4% | 75 | 2026-08-02 | independent | Source ↗ |
| Humanity's Last Exam | Active | 39.1% | 93 | 2026-08-02 | independent | Source ↗ |
| AIME | Nearing Saturation | 80.8% | 75 | 2026-08-02 | independent | Source ↗ |
| OlympiadBench | Active | 69.8% | 80 | 2026-08-01 | independent | Source ↗ |
| Terminal-Bench | Active | 60.8% | 80 | 2026-07-31 | independent | Source ↗ |
| FireBench | Active | 67.6% | 95 | 2026-07-31 | independent | Source ↗ |
| MathArena | Active | 68.1% | 99 | 2026-07-31 | independent | Source ↗ |
| SimpleQA | Active | 47.6% | 90 | 2026-07-30 | independent | Source ↗ |
| OpenAI-MRCR | Active | 76.9% | 80 | 2026-07-30 | independent | Source ↗ |
| CodeContests | Deprecated | 56.5% | 10 | 2026-07-30 | independent | Source ↗ |
| SWE-bench Pro | Active | 53.9% | 90 | 2026-07-29 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 66% | 80 | 2026-07-29 | independent | Source ↗ |
| LongBench-Pro | Active | 75% | 80 | 2026-07-29 | independent | Source ↗ |
| MultiPL-E | Deprecated | 79.9% | 10 | 2026-07-28 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 73.2% | 78 | 2026-07-28 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 77.8% | 65 | 2026-07-27 | independent | Source ↗ |
| LiveCodeBench | Saturated | 71.3% | 39 | 2026-07-27 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 54.7% | 80 | 2026-07-27 | independent | Source ↗ |
| MATH | Saturated | 83.4% | 25 | 2026-07-27 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 73.2% | 65 | 2026-07-26 | independent | Source ↗ |
| Mercury | Deprecated | 76.5% | 10 | 2026-07-26 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 84.2% | 75 | 2026-07-26 | independent | Source ↗ |
| TabMWP | Saturated | 78.3% | 25 | 2026-07-25 | independent | Source ↗ |
| CyberSecEval | Deprecated | 78.2% | 10 | 2026-07-25 | independent | Source ↗ |
| RULER | Nearing Saturation | 84.2% | 84 | 2026-07-25 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 67.8% | 25 | 2026-07-24 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 80.5% | 78 | 2026-07-24 | independent | Source ↗ |
| MMLU-Redux | Saturated | 89.7% | 25 | 2026-07-24 | independent | Source ↗ |
| SWE-bench | Deprecated | 59.1% | 10 | 2026-07-23 | independent | Source ↗ |
| MGSM | Nearing Saturation | 80.8% | 65 | 2026-07-23 | independent | Source ↗ |
| LongBench | Nearing Saturation | 78.7% | 65 | 2026-07-23 | independent | Source ↗ |
| QASPER | Nearing Saturation | 75.9% | 65 | 2026-07-22 | independent | Source ↗ |
| TruthfulQA | Saturated | 80.5% | 25 | 2026-07-22 | independent | Source ↗ |
| AGIEval | Saturated | 76.4% | 25 | 2026-07-22 | independent | Source ↗ |
| SuperGLUE | Saturated | 90.6% | 25 | 2026-07-21 | independent | Source ↗ |
| GSM-Hard | Deprecated | 81.7% | 10 | 2026-07-21 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 8.3% | 75 | 2026-07-20 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 87% | 25 | 2026-07-20 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 81.7% | 23 | 2026-07-20 | independent | Source ↗ |
| WinoGrande | Saturated | 92.4% | 28 | 2026-07-19 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 88.6% | 65 | 2026-07-19 | independent | Source ↗ |
| IFEval | Deprecated | 78.9% | 25 | 2026-07-19 | independent | Source ↗ |
| GPQA | Nearing Saturation | 80.5% | 78 | 2026-07-19 | independent | Source ↗ |
| HumanEval+ | Saturated | 93.9% | 25 | 2026-07-18 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 98.3% | 44 | 2026-07-18 | independent | Source ↗ |
| CommonsenseQA | Saturated | 93.2% | 28 | 2026-07-18 | independent | Source ↗ |
| GSM8K | Saturated | 91.9% | 25 | 2026-07-17 | independent | Source ↗ |
| GLUE | Saturated | 95.9% | 28 | 2026-07-17 | independent | Source ↗ |
| MBPP | Saturated | 92.1% | 25 | 2026-07-17 | independent | Source ↗ |
| HumanEval | Saturated | 99.1% | 25 | 2026-07-16 | independent | Source ↗ |
| MMLU | Saturated | 93.3% | 25 | 2026-07-16 | independent | Source ↗ |
| ARC-AGI | Saturated | 83.5% | 25 | 2026-07-16 | independent | Source ↗ |
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="moonshotai-kimi-k3",
messages=[{"role": "user", "content": "Evaluate multi-step logic problem"}],
temperature=0.2
)
print(response.choices[0].message.content)Evaluation & Sourcing Notes
Scores listed for MoonshotAI: Kimi K3 represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.