BacktestBench - Indicator Retrieval: leaderboard

The indicator retrieval stage of BacktestBench (2026), a quantitative-backtesting benchmark: accuracy of a model in extracting the trading factors and performance indicators a natural-language strategy refers to, assisted by BM25 lookup in a factor library. Percent; higher is better.

Metric: Accuracy (%) of the Summarizer stage in retrieving the factors and indicators a strategy uses (BM25-assisted), synthetic BacktestBench test split (the expert-crafted subset excluded), each model driving the AutoBacktest pipeline of the authors (Summarizer, SQL Retriever and Python Coder agents) over a PostgreSQL market database, temperature 0.6; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 23 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 3 Pro87.15#146
2Kimi K2 (Thinking)79.82
3Seed 1.876.93#258
4Qwen 3 235B A22B 2507 (Thinking)76.41#471
5Kimi Linear 48B A3B Instruct74.37#3027
6GLM-4.774.35#383
7Qwen 3 Next 80B A3B (Thinking)73.8#606
8Qwen 3 Max72.42#420
9Qwen 3 30B A3B 2507 (Thinking)69.73#769
10DeepSeek V3.268.98#425
11Seed-OSS-36B-InstructNon-reasoning68.07#855
12GPT-OSS-120B67.65#712
13GPT-OSS-20B67.54#1188
14Qwen 3 32B67.41#983
15Qwen 3 Coder Plus65.09#729

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=backtestbench-indicator-retrieval · How It Works · Data refreshed daily, snapshot 2026-11-22.