A · Anthropic
Fable 5
Preference #1 · Writing #1
Leads the vote
- Preference1507
- Code1627
- Writing1534
- Agents1511
Main event · 29 August 2026
Arena
Fable holds the vote. Opus holds the job.
Same prompt. Two models that actually lead. Score the artifact, not the speech.
B · Anthropic
Opus 5
Code #1 · Agents #1
Leads the code
- Preference1493
- Code1712
- Writing1490
- Agents1526
Leaderboard · 29 August 2026 · 73 models
The field.
Blind human votes. Who people pick when the names are hidden. Click a column or a row to put that model in the fight — up to five.
All 73 · Preference Elo
73 of 73 models. Preference from the public blind floor. Code from Arena Code. Writing and agents ranked on this desk. Bars are the spread inside the card, not Elo from zero.
The fight · 5 of 5
Line them up.
Two to five models. Four scores. Then run the same job on the floor below. Click a name on the chart to add or drop.
Fable 5
Anthropic
Opus 5
Anthropic
Muse 1.2
Meta
Qwen3.8 Max
Alibaba
Grok 4.6
xAI
Preference
Preference
1507
1493
1498
1479
1461
Code
Code
1627
1712
1535
1667
1631
Writing
Writing
1534
1490
1516
1488
1458
Agents
Agents
1511
1526
1472
1479
1466
The floor
Put them on the sand.
Same job. 5 models. Score the artifact.
A
B
C
D
E
A
Fable 5
Anthropic
Waiting on the floor.
B
Opus 5
Anthropic
Waiting on the floor.
C
Muse 1.2
Meta
Waiting on the floor.
D
Qwen3.8 Max
Alibaba
Waiting on the floor.
E
Grok 4.6
xAI
Waiting on the floor.
Vote unlocks after all answers
The method
How a round is scored.
Same job
Both models get the prompt you typed. No extra context, no tools on this floor — this is the prose and the code they would hand you.
The artifact
Vote for the one you would ship. A letter that sings and misses the date is still a miss. Tie if you cannot tell. Both miss if neither did the job.
A baseline
Last week’s agent, a script, or a careful human. If the new loop does not beat that, it does not leave the floor.
Next