GPT-4: benchmark results
OpenAI's GPT-4 flagship model (March 2023). Provider: OpenAI. Released 2023-03-14. Access: API.
Aggregate Rating 1601 ± 6, rank #438 of 1656 rated models, from 330 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AgentBoard | 70 | Progress Rate (imported) | 100 |
| BELLS | 91.7 | BELLS Score (%) | 100 |
| BizBench | 84.6 | Mean accuracy (%) over the eight BizBench tasks (FinKnow, Fo | 100 |
| BizBench - CodeFinQA | 78.8 | Accuracy (%; 844 FinQA questions over report text and tables | 100 |
| BizBench - CodeTAT-QA | 90.6 | Accuracy (%; 392 TAT-QA table questions answered by generati | 100 |
| BizBench - ConvFinQA Extract | 94 | Accuracy (%; 916 ConvFinQA questions answerable by extractin | 100 |
| BizBench - FinCode | 63.6 | Accuracy (%; 121 CFA and CPA exam questions answered by gene | 100 |
| BizBench - FinKnow | 80.1 | Accuracy (%; 877 multiple-choice business and finance knowle | 100 |
| BizBench - FormulaEval | 100 | Accuracy (%; 50 financial-formula functions to complete from | 100 |
| BizBench - SEC-Num | 79.3 | Accuracy (%; 2,000 SEC 10-K and 10-Q filing snippets, extrac | 100 |
| CanMT - Chinese to English | 5.52 | Translation score (1-7) on the 125 Chinese-to-English CanMT | 100 |
| CanMT - Chinese to Japanese | 5.34 | Translation score (1-7) on the 137 Chinese-to-Japanese CanMT | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-4 · How It Works · Data refreshed daily, snapshot 2026-10-20.