Task 6565 · Completed contest

Throughput, cost and quality, at the same time.

961 labelers and 7 frontier models estimated the calories in 3,490 meal photographs in 3 days.

402kOpinions collected115 reads per case, on average
$0.05Cost per caseAll reads and aggregation included
85.1 vs 74.0Consensus accuracy vs best model+11.1 points over GPT-5.2
Task-level quality

Every graded case, against ground truth

3,442 cases in this contest carry a known correct answer. This is where the accuracy-weighted consensus landed on all of them.

Consensus label ↓
Correct label →
0 to 150 cal150 to 300 calmore than 300 calRow total
0 to 150 cal1,20588.5% of column627.1%30.2%1,270
150 to 300 cal15211.2%69379.7% of column16713.8%1,012
more than 300 cal40.3%11513.2%1,04186.0% of column1,160
Cases with this answer1,3618701,2113,442
Consensus accuracy88.5%79.7%86.0%85.4%

Confusion matrix · Topic 6565 · 3,442 graded cases · 2,939 on the diagonal

One of the 1,041

A single case, from the highlighted cell

The matrix says the consensus is right 86% of the time on cases like this one. The replay shows the mechanism that gets it there.

Chicken and walnut salad on a plate
Case 33530108 · Image classification

How many calories do you think are in this meal?

A · 0 to 150 calB · 150 to 300 calC · more than 300 cal
25
qualified reads to settle
20
filtered below Q 70
80%
agreement at settle
real recorded reads

Contest statistics from Centaur task 6565; confusion matrix over its graded cases. Replay of real opinions from a live Centaur case; timeline compressed for playback, Q-scores as recorded at submission.