GPT-3.5: benchmark results
OpenAI's ChatGPT-era GPT-3.5 chat model family (gpt-3.5-turbo lineage), the low-cost API tier that powered ChatGPT's November 2022 launch. Provider: OpenAI. Released 2022-11-30. Access: API.
Aggregate Rating 1487 ± 11, rank #818 of 1656 rated models, from 63 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MHDash - Concern Type Accuracy | 45.33 | Concern-type accuracy (%, 7 classes) on the MHDash test spli | 100 |
| MHDash - Concern Type Macro Recall | 26.65 | Concern-type recall averaged over the 7 classes (%) on the M | 100 |
| MHDash - Joint Accuracy | 30 | Joint accuracy (%, both the concern type and the risk level | 100 |
| MHDash - Risk Level Accuracy | 35.33 | Risk-level accuracy (%, 6 classes) on the MHDash test split | 100 |
| BizBench - CodeFinQA | 67.5 | Accuracy (%; 844 FinQA questions over report text and tables | 93.3 |
| BizBench - CodeTAT-QA | 87.6 | Accuracy (%; 392 TAT-QA table questions answered by generati | 93.3 |
| BizBench - FinCode | 36.1 | Accuracy (%; 121 CFA and CPA exam questions answered by gene | 93.3 |
| BizBench - FinKnow | 60.3 | Accuracy (%; 877 multiple-choice business and finance knowle | 93.3 |
| BizBench | 72.5 | Mean accuracy (%) over the eight BizBench tasks (FinKnow, Fo | 92.9 |
| TerraLogic | 24.4 | Final answer accuracy (%, AnsAcc) over the 545 TerraLogic op | 90.9 |
| ClassEval | 29.6 | Class-Level Pass@1 (%) | 90 |
| SuperCLUE General (October 2023) - Open-Ended Questions (OPEN) | 73.12 | Score | 90 |
Interactive version: theaggregate.ai/model?slug=gpt-3-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.