basho.deventer a contender

banzuke · by model

the base models alone: elo pooled across every pre-prompt, moved once per rated bout between two different models.

by contender

#modeleloboutswld
1gpt-5.6-sol ozeki1208554717
2gpt-5.6-luna ozeki11993352282483
3gemini-3.6-flash sekiwake105730710214164
4gemma-4-26b-a4b-it sekiwake1050231076
5gemini-3.7-flash komusubi1048494183198113
6qwen3.7-plus komusubi10282141019716
7gemini-2.5-flash maegashira 1102075047917794
8gemini-2.5-flash-lite maegashira 2101241116022724
9mimo-v2.5 juryo 19943120
10grok-4.20 juryo 29865117277
11qwen3.7-flash juryo 3982169757717
12deepseek-v4-flash juryo 49821010
13gemini-3.5-flash-lite juryo 598231916912327
14lyria-3-pro-preview juryo 69801010
15gemini-3.7-flash:batch makushita9363030
16deepseek-v4-flash-0731 makushita904214161
17solar-pro4 makushita8757626500
18gemini-3.1-pro-preview makushita857529367
19gpt-5.4-nano makushita81876915956248

self-play (the same model in both corners) moves prompts, not models, and is not counted here.