Model Arena
Rankings show position. Benchmarks show strengths.
AA ranking; unified model profiles retain source-specific results.
Overall ranking · Top 10
Select a model to view both radars, using the same variant as this leaderboard.
Loading saved rankings…
Overall ranking · Top 30
A model and its closest contenders
The same models, test by test
Compare nearby models from the main board across individual tests. Adjacent ranks do not prove a significant difference.
The same models, test by test
Top 10Highest rank first, eight seconds per model. Capability and efficiency stay separate. Playback pauses in the background.
Music & artwork credits
“Cipher” — Kevin MacLeod (incompetech.com), 150 BPM. CC BY 4.0; Music sourceArtwork sources and authors appear on each model card.
See each model’s strengths
Colors show relative position among all comparable models. Top 10/20 does not change the scale. Missing values stay empty; select a value for evidence.
Faster or more affordable?
Price per million tokens on the horizontal axis; output speed on the vertical. Upper left is faster and cheaper. Only confirmed identities with matching observation dates are shown.
How scores change
Shows the four highest-ranked selected models. Dates are actual site snapshots. Missing points stay empty; history is never invented.
Beyond a single score
Integrated scores are available here. Other tests link to official methods; rankings appear only after verifiable data is connected.
Released; awaiting public evaluations
Data & updates
Manage sources ↗Sources rank independently. Missing scores are not zero. Radar values are 0–100 relative positions; raw scores and conditions are available.