Search-to-World - Request Resolution: leaderboard
Observed Retrieval Rate on 300 requests: percentage of runs retrieving at least one relevant direct 3D scene or corresponding video candidate. Each evaluated model fills all four primary agent roles; this metric precedes successful world delivery.
Metric: Observed Retrieval Rate (%). Source: arxiv.org. Saturation forecast: Not forecast. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 88.3 | #155 |
| 2 | GPT-5.6 Terra | 72.7 | #40 |
| 3 | Qwen 3.5 397B A17B | 68.7 | #154 |
| 4 | GPT-5.6 Sol | 68 | #22 |
| 5 | GLM-5.2 | 66 | #80 |
| 6 | DeepSeek V4 Pro | 63.7 | #112 |
| 7 | Kimi K3 | 54 | #35 |
| 8 | Qwen 3.5 35B A3B | 49.3 | #268 |
| 9 | Qwen 3.5 4B | 42.7 | #514 |
| 10 | Claude Fable 5 | 41 | #20 |
| 11 | Qwen 3.5 9B | 34 | #391 |
| 12 | Claude Opus 5 | 23.7 | #16 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=search-to-world-request-resolution · How It Works · Data refreshed daily, snapshot 2026-11-05.