PortBench-QA - Rebalancing: leaderboard
PortBench-QA's rebalancing template (Zhao, Chen and Su, 2026): from current and target weights the model decides whether to rebalance and, if so, names the most off-target asset and the trade direction, with partial credit for each part. Item score in percent over 50 test-period (2024-2025) questions, zero-shot at temperature 0 with thinking on; higher is better.
Metric: Item score (%; rebalance-or-hold decision with the corrective trade's asset and direction, partial credit, 50 test questions per model). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.1 (Thinking) | 88.2 | #175 (GLM-5.1) |
| 2 | Qwen 3.7 Max (Thinking) | 72.4 | #142 (Qwen 3.7 Max) |
| 3 | Kimi K2.6 (Thinking) | 68.4 | #173 (Kimi K2.6) |
| 4 | Qwen 3.6 Plus (Thinking) | 64 | #217 (Qwen 3.6 Plus) |
| 5 | Qwen 3.6 35B A3B (Thinking) | 56.4 | #461 (Qwen 3.6 35B A3B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=portbench-qa-rebalancing · How It Works · Data refreshed daily, snapshot 2026-11-12.