Preview unavailable for this view.
Model
Fable 5.1 vs Fable 5
100 website attempts.
Two models. No labels.
Choose the attempt you would actually use. Candidate identities and cutoff details appear only after your vote.
- Attempts
- 100
- Blind matchups
- 50
- Verdict layers
- 3
Loading the blind test
Manifest error
The benchmark data could not be loaded.
Check the manifest path and schema, then reload this page.
Builds in progress
The voting room is ready.
The comparison opens as soon as the generated manifest contains its first matchup.
Open demo data0 of 0 decisions
Votes stay on this device until a vote endpoint is connected.
Choose the stronger build
Judge what is visible and usable. Generation details and model identity stay hidden for now.
Preview unavailable for this view.
Model
Which result wins?
Keyboard: 1 for A, 2 for tie, 3 for B
Your vote is locked
You chose candidate A.
Audience totals will appear here when available.
Blinded AI judge
Judge result pending.
Three independent verdicts
Agreement matters more than a blended score.
Human preference, blinded AI judgment, and browser checks remain separate.
Complete every loaded matchup to reveal the model leaderboards.
0 decisions remainingSame brief. Fresh attempt. Hidden identity.
Every matchup compares the same prompt and attempt number. Side order is balanced and randomized for each browser.
When an attempt reaches the 16K limit, its preview is the exact truncated fragment returned at cutoff. Nothing is repaired before voting.
Candidate identity, completion, token counts, cost, timing, browser checks, and judge results appear only after that matchup's vote.