# Writing Composite Index

> Creative fiction and long-form narrative writing.

- Canonical URL: https://prochoice.industries/writing
- Domain JSON endpoint: https://prochoice.industries/api/v1/domains/writing
- Epoch version: `writing-v1.0.0`
- Snapshot hash: `43237d0f0a7497fc`
- Updated: 2026-11-21
- Cost axis: API cost per 10k-input / 4k-output writing task (USD, log scale)
- Evaluated entries: 88

## Headline picks

- Budget: GPT-6 Luna at 44.4 and $0.0030
- Best value: Claude Opus 5.5 at 74.6 and $0.12
- Top score: GPT-6.2 Astra at 79.2 and $0.24

## Pareto frontier

| Rank | Model | Effort | Provider | Score | Cost / task |
| :--- | :--- | :--- | :--- | :--- | :--- |
| 1 | GPT-6.2 Astra | High | OpenAI | 79.2 | $0.24 |
| 2 | Claude Opus 5.5 | High | Anthropic | 74.6 | $0.12 |
| 3 | Gemini 4 Carbon | High | Google | 67.8 | $0.060 |
| 4 | Gemini 4 Hydrogen | High | Google | 60.0 | $0.045 |
| 5 | Gemini 3.8 Flash | High | Google | 58.1 | $0.022 |
| 6 | GPT-5.6 Luna | Medium | OpenAI | 53.6 | $0.0068 |
| 7 | GPT-6 Luna | Max | OpenAI | 44.4 | $0.0030 |

## Active benchmarks

- LMArena Creative Writing (text_style_control, ACTIVE): https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset
- Lech Mazur Creative Writing (2026.11, ACTIVE): https://github.com/lechmazur/writing
- EQ-Bench Longform v3 (3.0, ACTIVE): https://eqbench.com/creative_writing_longform.html

Full JSON dataset: https://prochoice.industries/api/v1/snapshot
