> ## Documentation Index
> Fetch the complete documentation index at: https://docs.scaledown.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# ScaleDown vs OpenAI

> ScaleDown against GPT-5.5, GPT-5.4 Mini and GPT-5.4 Nano.

Compared against **GPT-5.5**, **GPT-5.4 Mini** and **GPT-5.4 Nano** on the four benchmark tasks.

## Accuracy

ScaleDown averages **8.72% higher accuracy** than the OpenAI average.

| Task | Dataset | ScaleDown vs OpenAI avg |
| - | - | - |
| Classify | CLINC OOS | +5.40% |
| Extract | CUAD | +14.40% |
| Summarize | QMSum | +8.83% |
| Compress | FinanceBench | +6.23% |

## Latency

ScaleDown is about **2.4x faster** than the OpenAI average. Response time in milliseconds:

| Task | ScaleDown | GPT-5.5 | Mini | Nano | OpenAI avg |
| - | - | - | - | - | - |
| Classify | 647 | 3,592 | 694 | 697 | 1,661 |
| Extract | 918 | 6,083 | 1,042 | 1,860 | 2,995 |
| Summarize | 1,426 | 7,805 | 1,856 | 1,759 | 3,807 |
| Compress | 2,658 | 8,217 | 3,087 | 3,426 | 4,910 |

[Cost and full charts](https://scaledown.ai/benchmarks/scaledown-vs-openai): Per-1K-request cost, accuracy charts and the cost-vs-accuracy plot on scaledown.ai.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.