When GPT5 comes out, lists its capabilities according to all quantifiable metrics, and compare that to the same for GPT4.
I will take the top 5 benchmarks/metrics that smart people I know tell me are the most credible/important, and that are available for both GPT's. Then I will calculate the percent increase/decrease from GPT4 to GPT5 for all five benchmarks, and average them.
This market resolves to whatever number that is.
Example of how I will calculate it:
Five benchmarks that each score between 0-100 points
(The actual real world benchmark scoring will surely vary, but I'll normalize them to a consistent scale first if OpenAI has not already done that for me in the GPT5 paper)
GPT4 scores an average score of 50
GPT5 scores an average score of 75
GPT5/GPT4 = 1.5
1.5 x 100 = 150, which is what this market would resolve to
This market is capped at 300
1,000
3.00The fact that this is phrased as "aggregate improvement over" rather than "aggregate score compared to" is odd. Improvement suggests you would subtract 100 from the value. If GPT5 scores 120 and GPT4 scores 80 on something, that is a 50% improvement, not 150. 150% is GPT5's score in comparison, when GPT4 = 100%.