Score aggregation
The five capabilities are equally weighted. Within each capability, metrics are averaged within datasets, then across datasets.
Source: Table 1 of the paper.← Back to the project
Paper-reported results · October 2026
IndiEval.
Individual simulation, evaluated in five dimensions.
This page presents the fixed evaluation reported in the paper; online submissions are not available.
F · O · N · T · S13Compared models
5Capabilities
29Native metrics
6,297Paper evaluation cases
Scroll horizontally to explore all capability scores →
| Rank | Model | Type |
|---|
Avg. uses the published score. Capability scores are recomputed from the rounded Table 1 metrics and shown to one decimal; small rounding differences may occur.
LOOK CLOSER / 29 NATIVE METRICS
Metric details
Select a model to explore its 29 native metrics.
Evaluation scope
The full evaluation comprises 13 benchmark entries and 6,297 cases. The public data release contains 12 subsets and 4,768 cases, excluding BehaviorChain.
Mistakes, HiToM and τ-USI are additional OOD tests and are excluded from Avg. here.