Rynn-GEU - Direct Answer - Position: leaderboard
RynnUnify Table 3 direct answering from original benchmark inputs, without generated-video rollouts or the additional RL-trained agent. Rynn-GEU has ten subsets; the native task-specific scores and their published average are reported on a 0–100 scale. Per-model effort/frame settings are not reported and remain source-scoped. Public oracle-rollout release is a different input protocol and does not substitute for these direct answers. Reported metric: Position.
Metric: Published task score (0–100). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Terra (Rynn-GEU Run: direct-answer; effort and frames unreported) | 44 | |
| 2 | Qwen3.5 9B (Rynn-GEU Run: direct-answer; effort and frames unreported) | 42 | |
| 3 | RynnUnify 16B (Rynn-GEU Run: direct-answer; effort and frames unreported) | 38.5 | |
| 4 | MiMo Embodied 7B (Rynn-GEU Run: direct-answer; effort and frames unreported) | 34 | |
| 5 | Claude Sonnet 5 (Rynn-GEU Run: direct-answer; effort and frames unreported) | 31.5 | |
| 6 | Embodied R1.5 8B (Rynn-GEU Run: direct-answer; effort and frames unreported) | 28 | |
| 7 | Qwen3 VL 8B Instruct (Rynn-GEU Run: direct-answer; effort and frames unreported) | 27.5 | |
| 8 | Cosmos 3 Nano (Rynn-GEU Run: direct-answer; effort and frames unreported) | 27 | |
| 9 | RxBrain (Rynn-GEU Run: direct-answer; effort and frames unreported) | 24.5 | |
| 10 | BAGEL (Rynn-GEU Run: direct-answer; effort and frames unreported) | 11.5 |
Interactive version: theaggregate.ai/benchmark?slug=rynn-geu-direct-answer-position · How It Works · Data refreshed daily, snapshot 2026-11-05.