How the rankings work

Version 8.0 of the AI Power Rankings ranks each kind of tool in its own list and scores each kind with its own weights.

Three lists

Models, coding harnesses and productivity tools are ranked in separate lists. A tool is ranked in one list at most, and its rank, tier and change are counted within that list. Ranks are not compared across lists.

The main rankings page shows the coding harnesses list.

Image and video models are not in the models list. Tools of other kinds are not ranked.

A list with fewer than 10 entries uses the tiers S, A and B by thirds and is not indexed by search engines.

Scoring profiles

Each list is scored with the profile for its kind. Every profile uses the same eight factors, and its weights add up to 100%, so every score is on the same 0-100 scale.

Coding harnesses

FactorWeight
Agentic capability12%
Innovation10%
Technical performance17%
Developer adoption17%
Market traction12%
Business sentiment12%
Development velocity12%
Platform resilience8%

Models

FactorWeight
Agentic capability18%
Innovation10%
Technical performance22%
Developer adoption10%
Market traction10%
Business sentiment12%
Development velocity10%
Platform resilience8%

Productivity tools

Provisional: the productivity tools list uses the coding-harness weights until its own scoring profile is decided.

FactorWeight
Agentic capability12%
Innovation10%
Technical performance17%
Developer adoption17%
Market traction12%
Business sentiment12%
Development velocity12%
Platform resilience8%

Whether a model's weights are open is not a scoring factor.

Ties

When two tools in a list have the same score, the one ranked higher in the last rankings published before version 8.0 comes first. A tool that was not ranked then comes after them.

Data sources

Scores use the metrics stored for each tool: benchmark results, adoption such as downloads, stars and users, funding and business news, and release activity. The rankings are recalculated once a month.