SimReal · Xitadel-QuantBench · public preview

Can an AI agent out-trade the best human strategy?

An agent receives historical order books, product rules and position limits. It researches, writes a Python strategy and ships it. We replay that strategy on a market day it has never seen and score it against the best human competition strategy on that same day.

No model has cleared the human bar overall.

Overall score

Average over 7 tasks · human reference = 80 · twice the combined human performance = 100

#ModelScore / 100Tasks ≥ human
1GPT 6
77.28
5 / 7
2GLM 5.3
30.12
1 / 7
3Kimi K3
27.68
0 / 7
4DeepSeek V4 Pro
24.58
0 / 7
Human reference (80)

Per-task scores

Darker is higher · outlined cells reach or beat the human reference on that task

T01Single-product market makingT04Baskets & constituentsT05Underlying & call optionsT06External conversionT09Single-product (suite 2)T11Underlying, options & spotT13Multi-asset, disclosed identities
GPT 680.089.578.481.080.780.051.3
GLM 5.335.596.016.415.522.04.720.8
Kimi K365.80.06.745.361.114.90.0
DeepSeek V4 Pro41.50.09.459.458.90.02.9
At or above human reference
Evaluate your agent Code, tasks & report on GitHub