Independent model evaluation · v4.1
Every frontier model,
one comparable number.
We run the same nine benchmark suites against every publicly available language model — reasoning, coding, agentic tool use, long-context recall — and normalise the results into a single Intelligence Index, alongside measured price and throughput.
Leaderboard
Click any row for the full benchmark breakdown. Sort by tapping a column header.
| # | Model | Index ▾ | Coding ▾ | Tok/s ▾ | $ / 1M ▾ | Context ▾ |
|---|
Intelligence vs. cost
The frontier is the set of models nothing beats on both axes at once. Everything above and to the left is a better deal.
Price–performance scatter
Blended $/1M tokens (log scale) against Intelligence Index. Hover a point for detail.
Top 10 by Intelligence
Follows the metric selected in the leaderboard controls.
Frontier over time
Best Intelligence Index available at any price, versus the best available under $1 per million blended tokens.
Methodology
Every score on this page is produced by our own harness. Nothing is copied from a vendor's release post.
Same harness, every model
All models run through one evaluation harness with identical prompting, identical tool schemas and no vendor-specific scaffolding. Reasoning models are given their default effort setting.
Measured, not quoted
Throughput and latency are median values sampled every fifteen minutes from the provider's public endpoint over the trailing seven days, from three geographic regions.
Normalised composite
The Intelligence Index is the unweighted mean of nine suites, each rescaled so that a random baseline maps to zero. Suites rotate out once frontier saturation exceeds 95%.