ST-Bench (Image Input) - Spatio-Temporal Forecasting: leaderboard
In-context forecasting task of ST-Bench (Ni, Wang, He, Jin and Jin, STReasoner paper, 2026): given the history of a node graph's time series and a short context hint, the model predicts a node's next values. The model sees the node series rendered as plot images with the graph structure in text; scenarios are synthetic network simulations and Claude-4.5-Sonnet wrote the questions. The score is the mean absolute error of the forecast, in the scenario's own units, over 280 held-out questions; lower is better.
Metric: Mean absolute error of the forecast values (280 questions; lower is better). Source: arxiv.org. Saturation forecast: Around March 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 63.36 | #244 |
| 2 | GPT-5.2 | 64.7 | #172 |
| 3 | Qwen 3 VL 8B Instruct | 74.21 | #982 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=st-bench-image-input-spatio-temporal-forecasting · How It Works · Data refreshed daily, snapshot 2026-11-12.