Support

Benchmarks · Human preference

PriorsBench

Star0

What Priors visitors think: each ranking people finish on Priors counts every higher-placed model as beating every lower one, fitted into an Elo-scale rating like an arena, with recent rankings counting most. A measure of belief, not ability; never in a default mix.

Quality 71 of 100 · Private test set, with questions replaced on a schedule · Has measured 10 of today's 10 default lineup models

41 models · 41 results · updated 2026-11-22 · mypriors.com ↗

Leaderboard

Fable 5.2

Benchmark quality

How the score works
  • Overall
    71

    71 of 100

    How useful it is for telling today’s models apart

  • Headroom20%
    50

    Some room left

    Can top models still improve on it?

  • Test set20%
    100

    Private, refreshed

    Could labs train on the questions?

  • Recency15%
    100

    0 days old

    How new is it?

  • Grading15%
    40

    People vote

    How are answers scored?

  • New models15%
    100

    10 of 10 top models

    How fast does it test new releases?

  • Efforts10%
    0

    One effort per model

    Does it test each model at several reasoning efforts?

  • Breadth5%
    100

    41 models

    How many models has it measured?

How each part was worked out
Headroom 50 of 100 · 20% · +10
From the spread of published scores and the room left above the top quarter of models, one score per model.
Test set 100 of 100 · 20% · +20
Private test set, with questions replaced on a schedule. There is no test set: the votes are the rankings people finish on Priors, which keep arriving. Source
Recency 100 of 100 · 15% · +15
Released 2026-11-22 (from the publisher's pages, not yet checked by hand).
Grading 40 of 100 · 15% · +6
Ranked by people's votes.
New models 100 of 100 · 15% · +15
Has measured 10 of today's 10 default lineup models. Too few new models to time how quickly it adds them.
Efforts 0 of 100 · 10% · 0
Measured 0 of the 41 models that come in more than one reasoning effort at two or more of them.
Breadth 100 of 100 · 5% · +5
41 models measured.

Not counted in the score: 0 stars · used by 0 published pages. Both lift it in the directory’s recommended order. Test set, release date and grading were researched from the publisher’s pages (2026-11-19); tell us if one is wrong.

More from this benchmark

2 published measures

Breakdown

darker is better within each column · click a column to sort
Model
Fable 5.2354.0310.3
Gemini 4 Neon138.0131.5
Opus 5.51,487787.5
GPT-6.2 Astra167.0157.5
Gemini 4 Carbon765.0508.5
GPT-6.2 Sol168.0157.7
GPT-6 Astra1,474775.1
Sonnet 5.51,418755.5
GPT-6.1 Sol1,439773.5
Gemini 4 Hydrogen272.0243.2
GPT-6.1 Terra837.0549.2
Fable 5.11,432744.6
Gemini 4 Argon1,352719.8
Muse Wonder Full1,166637.0
Kimi K3.1839.0538.2
Fable 5656.0235.7
Opus 5969.0480.0
Muse Wonder1,030512.0
GPT-6 Sol1,124560.5
Gemini 3.7 Flash36.019.3
GPT-5.528.010.0
Muse Spark 1.31,150520.2
GPT-5.6 Sol675.0258.1
DeepSeek V4.1 Flash78.035.0
Opus 4.840.019.7

All results

41 rows
#ModelEffortPriorsBench
1Fable 5.22020 ±63
2Gemini 4 Neon1822 ±61
3Opus 5.51757 ±34
4GPT-6.2 Astra1718 ±59
5Gemini 4 Carbon1519 ±37
6GPT-6.2 Sol1501 ±35
7GPT-6 Astra1470 ±30
8Sonnet 5.51342 ±29
9GPT-6.1 Sol1302 ±29
10Gemini 4 Hydrogen1215 ±36
11GPT-6.1 Terra1146 ±33
12Fable 5.11142 ±35
13Gemini 4 Argon1106 ±30
14Muse Wonder Full1087 ±32
15Kimi K3.1895 ±38
16Fable 5804 ±36
17Opus 5776 ±39
18Muse Wonder711 ±39
19GPT-6 Sol711 ±40
20Gemini 3.7 Flash636 ±177
21GPT-5.5608 ±87
22Muse Spark 1.3545 ±49
23GPT-5.6 Sol505 ±50
24DeepSeek V4.1 Flash499 ±93
25Opus 4.8474 ±118
26Qwen 4 Max455 ±115
27Gemini 3.8 Flash432 ±125
28Haiku 5.5427 ±58
29GLM 5.3 Flash416 ±131
30Grok 4.7367 ±56