Kimi Linear 48B A3B Instruct: benchmark results
Moonshot's 48B MoE (3B active) with hybrid Kimi Delta linear attention and a 1M context, MIT-licensed. Provider: Moonshot. Released 2025-10-30. Access: Open.
Unified ELO 1495 ± 1, rank #714 of 1392 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| UGI - Writing | 37.74 | Writing Score | 64.8 |
| UGI - Natural Intelligence | 22.09 | NatInt Score | 51.2 |
| AA Terminal-Bench Hard | 11.36 | Accuracy (%) | 46.8 |
| SpeechMap Compliance | 53.4 | % Requests Completed | 37.2 |
| AA AIME 2025 | 36.33 | Accuracy (%) | 33.7 |
| AA LiveCodeBench | 37.78 | Pass@1 (%) | 33.3 |
| AA Long Context Reasoning | 28 | Accuracy (%) | 30.9 |
| BenchmarkList ECI | 94.51 | Capability Index (ECI) | 27.2 |
| Artificial Analysis Intelligence Index | 2.9 | Intelligence Index | 26.4 |
| ATLAS | 33.1 | ATLAScore (self-reported) | 22.7 |
| UGI - Willingness (W/10) | 2.8 | W/10 Score | 20.6 |
| SvelteBench | 66.7 | Average pass@1 (%) | 19.2 |
Interactive version: theaggregate.ai/model?slug=kimi-linear-48b-a3b-instruct · How It Works · Data refreshed daily, snapshot 2026-09-05.