QuantCodeEval (Claude Code): leaderboard
QuantCodeEval — reproducing quantitative-finance strategies from 30 research papers, run under the Claude Code agent harness. Full Pass Rate (%) counts an attempt only when every checker in the task contract passes: 572 property checkers verify each pipeline stage (data loading, features, model fitting, signal, portfolio, execution), so matching the paper's headline Sharpe is not enough.
Metric: Full Pass Rate (%). Source: www.quantcodeeval.cloud. Saturation forecast: Around 2029. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 31.33 | #16 |
| 2 | Claude Opus 4.8 | 27.33 | #39 |
| 3 | Claude Opus 4.7 | 24 | #51 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=quantcodeeval-claude-code · How It Works · Data refreshed daily, snapshot 2026-11-05.