Lyra 2 Research & craft · October 2026
Built for the
whole story.
A benchmark-leading writing foundation, guided by your ideas and engineered to carry them through a complete book.
Start my book Explore the evidence01 / Writing quality
Good writing starts with a strong foundation.
We chose a foundation that leads the overall English writing ranking, then built Lyra around the work of making books.
English writing, overall
source ↗Top four published scores · higher is better
AI Writing Benchmark · 27 Oct 2026. Preference scores, not percentage grades. These evaluate the underlying writing models, not complete Lyra books.
Selected writing strengths
source ↗Our foundation’s category scores · higher is better
Selected category leaders in the same benchmark. Results vary by task; Grok 4.7 leads fiction and narrative.
The benchmark’s overall evidence base, with names hidden during judging. Human votes are tracked separately. method ↗
02 / The book engine
Every chapter belongs to the same book.
Lyra surrounds the writer with your brief, a book-wide plan and the context of the story already told.
- 01
Your direction
Genre, voice, point of view, source material and the details you want to protect.
- 02
A book-wide plan
Character dossiers, chapter purposes and a narrative arc before the prose begins.
- 03
Connected chapters
Accepted prose and continuity notes guide what happens next. Series memory carries the world forward.
- 04
Your finished draft
A complete manuscript and original cover, ready to read, edit and download.
Designed to preserve your intent across the manuscript. You make the final editorial decisions. Read real books ↗
03 / Prose in context
Strong prose. In more than one language.
A separate book-writing benchmark places our foundation among the leaders for German prose and nonfiction.
Book-writing benchmark results
source ↗Our foundation Comparison model
German literary prosevs. Claude Opus 5.5
1,382vs. 1,325
German nonfictionvs. Claude Opus 5.5
1,266vs. 1,226
English controlvs. Claude Sonnet 5.5
1,469vs. 1,195
Dots show ratings; lines show 95% confidence intervals. The German intervals overlap, so their order is not established. These Elo-like ratings use a different scale from the English scores above.
Study details and exact values
hermes3000 Writing Benchmark, HybridAI GmbH · 18 October 2026. Seven models, 294 texts, three AI judges. The study tested our foundation at medium reasoning; Lyra 2 uses high. Short and medium writing tasks do not establish full-book performance. Excluding own-family judges reverses the two German leaders. Human evidence is exploratory: one reader, 42 votes.
| Track | Our foundation | Comparison |
|---|---|---|
| German prose | 1,382 (1,320–1,453) | Claude Opus 5.5 1,325 (1,264–1,401) |
| German nonfiction | 1,266 (1,195–1,355) | Claude Opus 5.5 1,226 (1,172–1,287) |
| English control | 1,469 (1,354–1,660) | Claude Sonnet 5.5 1,195 (1,073–1,353) |
Selected prose tracks, not an all-task lead. The same foundation scored 80% in copy-editing and 98% on context assertions. Read the methodology and complete results. Adapted from the published results under CC BY 4.0.
04 / Practical by design
More of your ideas can become books.
The same Lyra writing engine on every plan. Choose the monthly capacity that fits your bookshelf.
A complete 40,000-word draft.
Planning, chapters and an original cover. Word, EPUB and Markdown downloads on paid plans.
8 creditswith standard Lyra
Find your plan ↗| Plan / month | Books | Per book |
|---|---|---|
| Author $24.99 / month | 1 | $24.99 |
| Publisher $59.99 / month | 3 | $20.00 |
| Mogul $179.99 / month | 12 | $15.00 |
| Disruptor $799.99 / month | 60 | $13.33 |
Monthly subscription price divided by books covered, assuming all monthly credits go to this length with standard Lyra. Rounded to cents. Full plan payment applies; these are not standalone book prices. Taxes may apply. pricing ↗
Evidence notes
The numbers, with their context.
Compiled by ImagineYourBook for Lyra 2. Benchmark snapshot: 27 October 2026.
English writing
AI Writing Benchmark compares anonymous responses using AI judges. Scores estimate preference against the average model, rather than correctness or a percentage quality grade. Our foundation uses the same high-reasoning setting as the ranked entry. The overall lead is a point estimate, not proof of superiority in every genre.
German prose and English control
hermes3000 Writing Benchmark, maintained by HybridAI GmbH, maker of a writing app. Public prompts, outputs and judging methods make the results inspectable. The charts adapt its selected prose results, with confidence intervals and the original source linked above.
What belongs to Lyra
The benchmark scores describe Lyra’s underlying writing foundation. Our planning, context management, continuity workflow and book packaging are additional engineering. A separately scored evaluation of the complete Lyra pipeline is not reported here. Read finished books to see that workflow in practice.
Cost per book
Calculated from our published monthly plans and standard Lyra’s one credit per 5,000 target words. Annual billing, pay-once credits, promotions and Lyra Max have different economics. Writing length is a target; review your finished manuscript before publishing.
The next story is yours.
Give it a world. Give it a voice.
Start my book