Understand how composite scores are built from individual signals. Each signal measures a distinct quality dimension, weighted and combined to produce the final score for every model.
High-level summary of the signal data across all scored models.
Signals Tracked
6
unique quality dimensions
Models with Signal Data
300
of 300 total models
Avg Signals per Model
5.4
signals per model on average
Signals ranked by their average weight in the composite score calculation. Higher weight means greater influence on the final score.
| Signal | Avg Weight | Avg Score |
|---|---|---|
Benchmarks | 30.0% | 69.5 |
Capabilities | 26.3% | 60.4 |
Pricing | 21.3% | 8.5 |
Recency | 15.0% | 76.2 |
Context Window | 13.1% | 81.6 |
Output Capacity | 13.1% | 59.2 |
Top 10 models by composite score with stacked signal contributions. Each colored segment is proportional to that signal's contribution to the total score.
For each signal, the top 5 models ranked by that signal's contribution to their composite score.
Which signals tend to move together? Pearson correlation coefficient between signal scores across all models. Values near +1 indicate signals that rise and fall together; values near -1 indicate inverse relationships.
How signals work and contribute to the composite score.
Signals are individual quality dimensions that capture different aspects of a model's value. Each signal measures a specific attribute such as benchmark performance, pricing efficiency, context capacity, or capability breadth. Together, they provide a multi-dimensional view of model quality.
Each signal is assigned a weight reflecting its importance in the overall assessment. Weights are expressed as fractions summing to 1.0 (100%). A signal with weight 0.25 contributes up to 25% of the composite score. Weights are calibrated based on the signal's relevance to practical model quality.
Each signal's raw value is normalized to a 0–100 scale to make signals comparable regardless of their original units. A score of 100 means the model ranks at the top for that signal, while 0 indicates the lowest possible performance. Z-scores are computed first, then mapped to the 0–100 range.
A signal's contribution equals its weight multiplied by its normalized score. For example, a signal with weight 0.25 and normalized score 80 contributes 20 points to the composite score. The sum of all contributions gives the final composite score. This makes it easy to see which signals drive each model's ranking.
Continue exploring AI model data with benchmarks, the full leaderboard, and other explorer views.
SignalScore breaks down into six components: Capability (breadth of supported features), Pricing (cost competitiveness), Context (input window size), Recency (how new the model is), Output (generation capacity), and Versatility (range of supported tasks and modalities).
Capability and Pricing each carry 25% weight, making them the two most impactful signals. A model that supports many capabilities (vision, function calling, streaming, reasoning) and has competitive pricing will score significantly higher than one that excels in only one dimension.
Some signals are positively correlated - models with large context windows tend to also have broad capabilities. Others show negative correlation - the most capable premium models often score low on pricing. Understanding these correlations helps explain why some models rank differently than expected.