Kimi K3: benchmark results
Moonshot AI's open-weight Kimi K3 (July 2026), a 2.8-trillion-parameter multimodal agentic MoE model built on Kimi Delta Attention, with native vision and a 1-million-token context for long-horizon coding and reasoning. Provider: Moonshot. Released 2026-07-16. Access: Open.
Aggregate Rating 1851 ± 3, rank #30 of 1656 rated models, from 1284 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AA Long Context Reasoning | 88.67 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - C | 90 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - Dart | 66 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - JavaScript | 78.18 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 68 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - Swift | 80 | Accuracy (%) | 100 |
| AA Omniscience - Software Engineering (SWE) - TypeScript | 77.78 | Accuracy (%) | 100 |
| AI Coding Daily (OpenCode) - Laravel Code Quality | 18.4 | Laravel Code Quality (max 20) points, LLM-judged rubric scor | 100 |
| APIFlow-Bench 1.0 - All Tasks | 89.85 | Pass Rate (%) | 100 |
| APIFlow-Bench 1.0 - Authentication | 89.7 | Pass Rate (%) | 100 |
| APIFlow-Bench 1.0 - Error Recovery | 90.3 | Pass Rate (%) | 100 |
| APIFlow-Bench 1.0 - Multi-step Workflow | 96.5 | Pass Rate (%) | 100 |
Interactive version: theaggregate.ai/model?slug=kimi-k3 · How It Works · Data refreshed daily, snapshot 2026-10-20.