# Model Intelligence profiles

**Model Intelligence** is a dedicated data portal (separate from the Model
Library) that provides benchmark scores, rankings, and detailed technical specs
for every discovered model. You can access it at
[/model-intelligence](/model-intelligence).

## What you can do

| Feature | What it provides |
|---|---|
| All Models | Browse 550+ model profiles with full benchmark data. Each profile page shows the model's scores across intelligence, speed, coding, and instruction-following benchmarks. |
| Leaderboards | Models ranked by intelligence score, request speed (tokens per second), latency, pricing, and context window size. |
| Providers | Aggregate data across 50+ providers — model counts, average scores, and price tiers. |
| Benchmarks | Dedicated per-benchmark pages with model rankings and comparison charts. |

## How it relates to the Model Library

The **Model Library** answers "what models are available" — it is a catalog
surface focused on provider, routing, and basic specs. **Model Intelligence**
answers "how good is this model" — it is a research surface with comparative
benchmark data.

Both surfaces are kept current from the same model discovery pipeline. The
Library is the day-to-day browser; Model Intelligence is the reference for
deeper evaluation.

## Freshness and limitations

Model Intelligence is refreshed when the model corpus is re-scraped. Key
limitations to be aware of:

- **Benchmark scores** are based on public third-party evaluations and
  provider-reported data. Scores may lag behind model updates.
- **Not all models have complete profiles.** A newly added model may appear
  in the Library before its benchmark data populates Intelligence.
- **Pricing signals** in the profiles are estimates. Always verify pricing
  against the provider's published rate sheet.
- **Leaderboard rankings** reflect the most recent scrape data, not real-time
  performance. Run your own evaluation for production-critical decisions.

## Opening a profile

From [/model-intelligence/models](/model-intelligence/models), click any model
to open its profile page. The profile page is a dynamic surface generated from
the scrape corpus — there is not a separate docs article per model.