Boards correlated
34
Spearman rho and Kendall tau-b correlations across quality boards, a clustered benchmark map, near-duplicate leaderboards and effective dimensionality.
34
358
4.41
41.7% variance on PC1
32
Spearman rho >= 0.90, n >= 10
Boards are grouped by how alike their model rankings are. Distance is sqrt(2 (1 - rho)) from Spearman rho, grouped by average linkage and cut at an average rho of about 0.7. Pairs need at least 10 shared models; the others are left blank. Agreeing rankings show that the same models lead, not that the tasks are the same.
Hover or use arrow keys to read a pair.
33 of 33 boards shown. Lines are the tree of closest pairs. The layout is approximate: stress 0.32 (lower is more faithful). The axes have no meaning. Points that are near each other have alike model rankings, not alike tasks.
| Board | Cluster | Nearest neighbor | Rho |
|---|---|---|---|
| LMArena Text (style control) | Cluster 1 | LMArena Text | 0.98 |
| LMArena Text | Cluster 1 | LMArena Text (style control) | 0.98 |
| LMArena Vision (style control) | Cluster 1 | LMArena Vision | 0.98 |
| LMArena Vision | Cluster 1 | LMArena Vision (style control) | 0.98 |
| GPQA diamond | Cluster 1 | METR time horizon | 0.95 |
| METR time horizon | Cluster 1 | GPQA diamond | 0.95 |
| FrontierMath-2025-02-28-Private | Cluster 1 | GPQA diamond | 0.94 |
| HLE | Cluster 1 | GPQA diamond | 0.95 |
| ARC-AGI-2 | Cluster 1 | GPQA diamond | 0.93 |
| Terminal Bench | Cluster 1 | HLE | 0.91 |
| SWE-Bench verified | Cluster 1 | ARC-AGI-2 | 0.91 |
| LMArena Search | Cluster 1 | METR time horizon | 0.92 |
| Bug Hunt Bench | Cluster 2 | Vals Index | 0.87 |
| LMArena Search (style control) | Cluster 2 | ARC-AGI-2 | 0.87 |
| Vending-Bench 2 | Cluster 2 | LMArena WebDev | 0.85 |
| LMArena Document (style control) | Cluster 3 | LMArena Document | 0.92 |
| LMArena Document | Cluster 3 | LMArena Document (style control) | 0.92 |
| Vals Index | Cluster 4 | Terminal-Bench 4 | 0.94 |
| Terminal-Bench 4 | Cluster 4 | Vals Index | 0.94 |
| Code Migration | Cluster 4 | Vals Index | 0.93 |
| LMArena Agent | Cluster 4 | LMArena WebDev | 0.90 |
| LMArena WebDev | Cluster 4 | LMArena Agent | 0.90 |
| CursorBench | Cluster 4 | EMB | 0.89 |
| EMB | Cluster 4 | GPQA diamond | 0.90 |
| FrontierCode | Cluster 4 | LMArena Agent | 0.87 |
| Vibe Code Bench | Cluster 4 | Vals Index | 0.85 |
| Legal Research | Cluster 5 | Tax Agent | 0.89 |
| Tax Agent | Cluster 5 | Legal Research | 0.89 |
| Finance Agent | Cluster 5 | CursorBench | 0.79 |
| Legal Agent Benchmark (HLAB-Vals) | On its own | Tax Agent | 0.66 |
| GDPval | On its own | ARC-AGI-2 | 0.79 |
| LMArena Image Edit | Cluster 6 | LMArena Text-to-Image | 0.93 |
| LMArena Text-to-Image | Cluster 6 | LMArena Image Edit | 0.93 |
Method rules
Known limits
Exclusions
| Board A | Board B | Shared models | Spearman rho | Kendall tau-b |
|---|---|---|---|---|
| lmarena-text-style-control/lmarena#overall | lmarena-text/lmarena#overall | 392 | 0.984 | 0.899 |
| lmarena-vision-style-control/lmarena#overall | lmarena-vision/lmarena#overall | 153 | 0.983 | 0.893 |
| lmarena-text-style-control/lmarena#overall | lmarena-vision-style-control/lmarena#overall | 125 | 0.978 | 0.871 |
| lmarena-text/lmarena#overall | lmarena-vision/lmarena#overall | 125 | 0.968 | 0.860 |
| lmarena-text-style-control/lmarena#overall | lmarena-vision/lmarena#overall | 125 | 0.957 | 0.827 |
| gpqa-diamond/epoch | metr-time-horizon/metr@1.1 | 20 | 0.952 | 0.832 |
| gpqa-diamond/epoch | hle/epoch | 36 | 0.950 | 0.828 |
| lmarena-text-style-control/lmarena#overall | metr-time-horizon/metr@1.1 | 20 | 0.950 | 0.821 |
| lmarena-vision/lmarena#overall | metr-time-horizon/metr@1.1 | 15 | 0.950 | 0.829 |
| lmarena-text/lmarena#overall | metr-time-horizon/metr@1.1 | 20 | 0.947 | 0.811 |
| lmarena-vision-style-control/lmarena#overall | metr-time-horizon/metr@1.1 | 15 | 0.946 | 0.810 |
| gpqa-diamond/epoch | lmarena-text-style-control/lmarena#overall | 160 | 0.943 | 0.792 |
| vals-index/vals | vals-terminal-bench-4/vals | 38 | 0.942 | 0.821 |
| frontiermath/epoch | gpqa-diamond/epoch | 67 | 0.935 | 0.787 |
| lmarena-text/lmarena#overall | lmarena-vision-style-control/lmarena#overall | 125 | 0.935 | 0.786 |
| frontiermath/epoch | hle/epoch | 26 | 0.934 | 0.793 |
| arc-agi-2/epoch | gpqa-diamond/epoch | 78 | 0.931 | 0.785 |
| lmarena-image-edit/lmarena#overall | lmarena-text-to-image/lmarena#overall | 43 | 0.930 | 0.790 |
| vals-code-migration/vals | vals-index/vals | 38 | 0.930 | 0.784 |
| arc-agi-2/epoch | hle/epoch | 31 | 0.927 | 0.797 |
| arc-agi-2/epoch | metr-time-horizon/metr@1.1 | 10 | 0.927 | 0.822 |
| gpqa-diamond/epoch | lmarena-text/lmarena#overall | 160 | 0.927 | 0.772 |
| vals-code-migration/vals | vals-terminal-bench-4/vals | 38 | 0.924 | 0.779 |
| frontiermath/epoch | lmarena-vision/lmarena#overall | 39 | 0.918 | 0.759 |
| lmarena-document-style-control/lmarena | lmarena-document/lmarena | 40 | 0.915 | 0.759 |
| lmarena-search/lmarena | metr-time-horizon/metr@1.1 | 10 | 0.915 | 0.778 |
| frontiermath/epoch | metr-time-horizon/metr@1.1 | 14 | 0.909 | 0.773 |
| arc-agi-2/epoch | swe-bench-verified/epoch | 22 | 0.907 | 0.761 |
| hle/epoch | terminal-bench/epoch | 13 | 0.905 | 0.761 |
| lmarena-agent/lmarena | lmarena-webdev/lmarena#overall | 55 | 0.904 | 0.752 |
| gpqa-diamond/epoch | vals-emb/vals | 22 | 0.903 | 0.743 |
| hle/epoch | metr-time-horizon/metr@1.1 | 13 | 0.901 | 0.795 |
| Domain | Boards | Within-domain pairs | Mean Spearman rho |
|---|---|---|---|
| Agentic | 5 | 4 | 0.574 |
| Coding | 9 | 24 | 0.793 |
| Finance | 4 | 3 | 0.655 |
| General text | 5 | 10 | 0.790 |
| Image generation | 2 | 1 | 0.930 |
| Legal | 2 | 1 | 0.641 |
| Math | 1 | 0 | Insufficient pairs |
| Science | 2 | 1 | 0.950 |
| Search | 2 | 1 | 0.811 |
| Vision | 2 | 1 | 0.983 |
Method rules
Known limits
Exclusions