QuantCode-Bench: leaderboard
Metric: Judge pass rate (%), single turn: share of the tasks whose generated strategy compiles, backtests without runtime errors, places at least one trade and is judged by an LLM to implement the described strategy, on QuantCode-Bench (400 Backtrader trading-strategy tasks from Reddit, TradingView, StackExchange, GitHub and synthetic sources, multi-timeframe historical data); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 75.8 |
| 2 | GPT-5.4 | 70.2 |
| 3 | Claude Sonnet 4.5 | 69.8 |
| 4 | GPT-5.2 Codex | 67.5 |
| 5 | GLM-5 | 65.4 |
| 6 | Claude Sonnet 4.6 | 65 |
| 7 | Kimi K2.5 | 64.8 |
| 8 | Gemini 3 Flash | 59.8 |
| 9 | Grok 4.1 Fast | 48.9 |
| 10 | DeepSeek V3.2 | 48.8 |
| 11 | Qwen 3 235B A22B | 48.2 |
| 12 | Qwen 3 Coder 30B A3B Instruct | 39.2 |
| 13 | Gemini 2.5 Flash | 31.2 |
| 14 | Qwen 3 14B | 25.2 |
| 15 | Qwen 3 8B | 18.5 |
Interactive version: theaggregate.ai/benchmark?slug=quantcode-bench · How It Works · Data refreshed daily, snapshot 2026-10-07.