Creative Writing with Claude & GPT: The Advanced Model Playbook
Route every fiction task to the right model, effort setting, and surface — Fable 5, Sonnet 5, Opus, GPT-5.6 Sol. Voice profiles, dialogue craft, lore bibles, de-AI-ing prose, and writing in Polish.
Requested by Vizz
You’ve noticed it: ask three different models for the same scene and you get three different writers. One nails the dialogue but loses the plot. One reasons brilliantly and writes like a press release. One is free and quietly outperforms the flagship you’re paying for.
This course is about exploiting that. The advanced writer’s edge in 2026 isn’t a better prompt — it’s routing: matching each task to the right model, the right effort setting, and the right surface, then killing the residual AI-tells at edit time with a disciplined revision pipeline.
The evidence says the obvious picks are wrong. The newest Opus regressed on creative writing versus its predecessors. ChatGPT’s free default beats its paid flagship in blind fiction tests. And a structured voice profile moves your output quality more than any model upgrade — a controlled test found a mid-tier model with a profile matching the top model without one.
You’ll build the whole pipeline on your own project: a routing table, a living voice profile, per-character dialogue cards, a lore bible engineered around real context limits, and a three-pass de-AI revision system — plus a full lesson on writing fiction directly in Polish without the translated feel.
Requested by a Pro member. Based on primary-source research (July 2026): Anthropic and OpenAI model documentation and release notes, creative-writing evals (EQ-Bench Creative, LMArena) with their error bars and judge-bias caveats, peer-reviewed work on AI-prose stylometry and translationese, and the workflows fiction writers actually share.
What You'll Learn
- Decode the full July 2026 lineup — Claude Fable 5, Opus 4.8, Sonnet 5, GPT-5.6 (Sol/Terra/Luna), GPT-5.5 — and what each is actually for
- Route each writing task to the right model, effort setting, and surface (chat, Projects, Claude Code, Cowork, Codex, API)
- Defend an evidence-based model verdict for fiction — including why the newest model is often the wrong pick for prose
- Extract a reusable voice profile from your own writing and attach it to every generation
- Engineer dialogue with per-character voice cards, subtext, and interruption patterns
- Architect book-length continuity: lore bibles, rolling summaries, and workflows that survive context rot
- Run a three-pass de-AI revision pipeline that kills AI-isms without sanding off your own style
- Write publishable fiction directly in Polish (or any non-English language) without the translated feel
After This Course, You Can
What You'll Build
Course Syllabus
Prerequisites
- Fluency with at least one frontier AI assistant (ChatGPT or Claude) — this course spends zero time on basic prompting
- An active writing project: a story, novel, narrative series, character bios, or fandom-wiki/lore content
- Comfort with technical settings (model pickers, API concepts, project files) — developer experience helps but isn't required
