# Little Dorrit Editor Benchmark

> A benchmark of vision language models interpreting handwritten editorial corrections on printed pages from Little Dorrit.

Use the dated results and exact model routes. Higher F1 is better, but small differences may not be meaningful. Missing prices or historical usage mean unknown, not zero. Token-price comparisons are not measured cost per page. OpenRouter and direct API entries are distinct benchmark routes.

## Benchmark

- [Overview and methodology](https://dorrit.pairsys.ai/index.md): Task, scoring, view definitions, and source links.
- [Rankings and token prices](https://dorrit.pairsys.ai/rankings.md): All models, dated prices, and preprocessing caveats.
- [Raw results](https://dorrit.pairsys.ai/results.json): Edit-level evaluated results.
- [Pricing snapshot](https://dorrit.pairsys.ai/pricing.json): Exact routes, source URLs, and verification dates.

## Optional

- [Technical report](https://dorrit.pairsys.ai/technical-report.pdf): Detailed benchmark methodology.
- [Repository](https://github.com/PAIR-Systems-Inc/little-dorrit-editor): Code and experiment configurations.
- [Dataset](https://huggingface.co/datasets/pairsys/little-dorrit-editor): Annotated benchmark data.