ReferFlow-Bench - Reset - T-Acc: leaderboard
Table 1; trajectory-level referring expression grounding; Q-Acc individual query, T-Acc full trajectory, F-Acc final query; shared benchmark, native model sizes retained; no general-REC mirror cells or training ablations. Metric: T-Acc (%). Higher is better. Source: https://arxiv.org/html/2610.20352v1
Metric: T-Acc (%). Source: arxiv.org. Saturation forecast: Not forecast. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.8 27B | 10.5 | #96 |
| 2 | Qwen 3.5 122B A10B | 10.2 | #169 |
| 3 | Qwen 3.5 27B | 9.7 | #183 |
| 4 | GLM-4.6V | 9.5 | #309 |
| 5 | Step 3.7 Flash | 9.4 | #231 |
| 6 | Qwen 3.5 9B | 8.9 | #367 |
| 7 | GPT-5.6 Sol | 8.4 | #18 |
| 8 | Qwen 3.5 4B | 8.2 | #463 |
| 9 | Gemini 3.1 Pro (Preview) | 5.4 | #56 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=referflow-bench-reset-t-acc · How It Works · Data refreshed daily, snapshot 2026-10-16.