Kimi K3 Tech Report - Agentic Vision Bench: leaderboard
Moonshot AI's in-house Agentic Vision Bench tests whether agents notice and correctly use key visual facts during task execution; all models run in the Kimi Code harness. Kimi K3 technical report, Table 3; score in percent, higher is better; models at maximum reasoning effort (GPT-5.5 at xhigh).
Metric: Score (%). Source: raw.githubusercontent.com. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 82.9 | #22 (GPT-5.6 Sol) |
| 2 | Claude Opus 4.8 (Max) | 82.8 | #39 (Claude Opus 4.8) |
| 3 | Claude Fable 5 (Max) | 81.1 | #20 (Claude Fable 5) |
| 4 | Kimi K3 (Max) | 78.3 | #35 (Kimi K3) |
| 5 | GPT-5.5 (xHigh) | 76.9 | #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-agentic-vision-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.