Support

PriorsBench

The models people rank highest on Priors.

rankings
1,519
head-to-head votes
150,194
models rated
41
first ranking
Sep 22
Add your ranking
  1. 2Gemini 4 Neon1,822±47
  2. 1Fable 5.22,019±57
  3. 3Opus 5.51,758±33

Leaderboard

#ModelRatingRank rangeRankings
1Fable 5.22,0191353
2newGemini 4 Neon1,8222137
31Opus 5.51,7583-41,486
4newGPT-6.2 Astra1,7173-4166
52Gemini 4 Carbon1,5195-7764
6newGPT-6.2 Sol1,5015-7167
73GPT-6 Astra1,4705-71,473
83Sonnet 5.51,3428-91,417
93GPT-6.1 Sol1,3028-91,438
103Gemini 4 Hydrogen1,21610271
112GPT-6.1 Terra1,14711-14836
124Fable 5.11,14311-141,431
133Gemini 4 Argon1,10711-141,351
143Muse Wonder Full1,08711-141,165
153Kimi K3.189615838

Head to head

Share of rankings where the row model is placed above the column model

Fable 5.2848494949697989898Gemini 4 Neon166858839396989798Opus 5.5163252689390979698GPT-6.2 Astra64248779696909495Gemini 4 Carbon61732237758736796GPT-6.2 Sol47742376829692GPT-6 Astra341044224718392Sonnet 5.5223102718295689GPT-6.1 Sol2346334174484Gemini 4 Hydrogen22254881116

Closest calls

Opus 5.5overGPT-6.2 Astra52%
166 rankings
Sonnet 5.5overGPT-6.1 Sol56%
1,218 rankings
Gemini 4 NeonoverGPT-6.2 Astra58%
132 rankings
Gemini 4 CarbonoverGPT-6 Astra58%
681 rankings
Gemini 4 CarbonoverGPT-6.1 Sol67%
682 rankings
How it’s scored
  • Each ranking counts every higher model as beating every lower one, fitted with Bradley-Terry on the Elo scale (400 points apart means placed higher 10 times in 11).
  • Recent rankings count more: weight halves every 30 days. A model needs 15 rankings.
  • Rank range comes from 95% intervals over 100 resamples. A repeated lineup within a day counts once, and no day counts more than three times a typical day.
  • It measures belief, not ability, so it is never in a default mix. It is also on the benchmark list.