Kimi K3 Tech Report - Faithfulness: leaderboard
Moonshot AI's in-house Faithfulness evaluation measures factual hallucination in model responses, each response verified by a fact checker. Kimi K3 technical report, Table 3; reported as 1 minus the hallucination rate in percent, so higher is better. The harness column reads N/A.
Metric: 1 - Hallucination Rate (%). Source: raw.githubusercontent.com. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (xHigh) | 86.5 | #33 (GPT-5.5) |
| 2 | Kimi K3 (Max) | 85.5 | #35 (Kimi K3) |
| 3 | GPT-5.6 Sol (Max) | 84.8 | #22 (GPT-5.6 Sol) |
| 4 | Claude Opus 4.8 (Max) | 83.6 | #39 (Claude Opus 4.8) |
| 5 | GLM-5.2 (Max) | 74.8 | #80 (GLM-5.2) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kimi-k3-tech-report-faithfulness · How It Works · Data refreshed daily, snapshot 2026-11-05.