Skip to content
Math · ballpark

Ballpark

Twelve streams of calculation without tools (products, powers, long expressions, division, roots and logarithms, chained quotients, exponentials, real powers, factorials, binomial coefficients, trigonometry and generalized harmonic sums), growing in size. The reply is a number or a range: an exact answer scores highest, a correct range scores by its precision, and a miss costs more the further off it is.

Source
Generated
Score
ModelAxes score, λ = 1
Ladder
structural · difficulty level · 1, 2, 3, 4, 5, 6, 7, 8
Items
2,200 items · 110 public
Dimensions
stream (12)
ScoreAccuracy95% interval
Rank#ModelScore
1GPT-6.1 SolScore 83.1, interval ±1.5, accuracy 95.0%83.1
2GPT-6 LunaScore 56.3, interval ±2.7, accuracy 80.7%56.3
Score ↑$/task · log ←↖ Better: higher score, lower cost$0.0010$0.00305060708090GPT-6 LunaGPT-6.1 Sol

Ladder

Accuracy by difficulty level rung for every model; the focused model shows its Wilson band and x50.

0%25%50%75%100%12345678difficulty levelGPT-6.1 SolGPT-6 Luna
Modelx50
GPT-6.1 SolAbove ladder
GPT-6 LunaAbove ladder

Outcomes

Correct 3,922 (89%)Incorrect 475 (11%)Abstained 0 (0%)Format failure 3 (0%)
Model
Outcomes
Correct
1
95%
GPT-6.1 Sol: 2,091 correct, 109 incorrect, 0 abstained, 0 format failures
2
83%
GPT-6 Luna: 1,831 correct, 366 incorrect, 0 abstained, 3 format failures

Slices · stream

Score per stream value for every model.

GPT-6.1 Sol
73.4
65.6
90.2
84.2
95.0
71.3
95.5
77.2
91.5
79.8
83.8
89.8
GPT-6 Luna
52.6
47.4
73.7
43.1
84.6
59.5
64.3
41.9
65.1
65.5
40.1
37.5