ShiJianBench: leaderboard
ShiJianBench (2026) places an LLM investment advisor in conversation with 27 simulated investors, a multi-agent simulator calibrated on 7,199 real users, while Chinese public-fund history from 2021 to 2026 replays, and scores each advisor against a matched rollout without advice. The total score combines investor outcomes, platform business value and GPT-4o-judged content quality under a hard compliance gate, on a 0-100 scale averaged over five runs; higher is better.
Metric: Advisor impact score (0-100): compliance-gated composite 0.7 x investor-outcome + 0.2 x business-value + 0.1 x content-quality scores, each a paired uplift over a no-advisor rollout clipped to 0-100, full horizon 2021-02-19 to 2026-01-31, over five matched runs of 27 simulated investors (risk profile x financial literacy x asset tier, weighted by 7,199 real users) in an LLM investor simulator replaying Chinese public-fund market traces, with GPT-4o judging advisor messages. Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 63.01 | #258 |
| 2 | Claude Sonnet 4.6 | 62.68 | #161 |
| 3 | GPT-5 | 61.77 | #167 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=shijianbench · How It Works · Data refreshed daily, snapshot 2026-11-13.