Kimi K3 Tech Report - MIRA Bench: leaderboard
Moonshot AI's in-house Multi-Agent Infra for Routing and Assignment (MIRA) Bench covers long-chain, multi-role, multi-system enterprise collaboration, including when to delegate to subagents; all models run in the internal out-of-distribution MIRA harness. Kimi K3 technical report, Table 3; score in percent, higher is better.
Metric: Score (%). Source: raw.githubusercontent.com. Saturation forecast: Around May 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 (Max) | 72.9 | #20 (Claude Fable 5) |
| 2 | Kimi K3 (Max) | 64.1 | #35 (Kimi K3) |
| 3 | GPT-5.6 Sol (Max) | 62.2 | #22 (GPT-5.6 Sol) |
| 4 | Claude Opus 4.8 (Max) | 59.8 | #39 (Claude Opus 4.8) |
| 5 | GPT-5.5 (xHigh) | 54.6 | #33 (GPT-5.5) |
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-mira-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.