Quick answer
Claude Fable 5 (model ID: claude-fable-5) is Anthropic's first model trained specifically for creative writing — fiction, dialogue, screenwriting, lyrics, narrative game design. It's not a replacement for Opus 4.8; it's a sibling. Cheaper, faster, and dramatically better at voice, characterisation, and emotional pacing than any general-purpose frontier model we've tested. Available now in claude.ai and via the API at $6/M input, $30/M output.
Anthropic has been quietly hinting at "specialised models" for two years. Fable 5 is the first one to actually ship. It uses the same underlying architecture as Opus 4.8 but with a completely different post-training pipeline — heavy weight on creative-writing datasets, a custom constitution focused on narrative craft instead of helpfulness, and a new evaluation suite scored by published authors and screenwriters rather than the usual factual benchmarks.
We spent a week running Fable 5 head-to-head with Opus 4.8 and GPT-5 on the same creative prompts — short stories, dialogue scenes, screenplay openings, song lyrics, game NPC monologues. Fable 5 won the blind test about 70% of the time. The gap isn't subtle.
What makes Fable 5 different in practice
- Voice consistency: across a 10,000-word story, character voices stay distinct and stable. Opus 4.8 drifts after about 3,000 words; Fable 5 holds the line.
- Show, don't tell: when asked for an emotional scene, Fable 5 writes sensory detail and behavioural cues. Opus 4.8 still defaults to naming emotions outright ("she felt sad").
- Dialogue rhythm: Fable 5 understands subtext. Characters talk around things, contradict themselves, leave things unsaid. Opus 4.8 writes characters who say exactly what they mean.
- Cliché immunity: ask both models for a romance opening and Opus 4.8 gives you "their eyes met across a crowded room." Fable 5 gives you something specific and weird.
- Pacing: Fable 5 understands that fiction needs uneven sentence lengths and quiet beats. Its prose has actual rhythm.
Pricing and access
Fable 5 is priced significantly below Opus 4.8 — $6/M input tokens (vs $12.50/M on Opus 4.8) and $30/M output (vs $75/M). That makes it the cheapest frontier-class creative-writing tool by a wide margin. It's available now in claude.ai for Pro subscribers (auto-toggled when you choose "Creative" mode), and via the API using model ID `claude-fable-5`. 200K context window. Standard streaming, tool use, and prompt caching are supported.
If you write fiction or marketing copy with a strong voice — try Fable 5 today. The cost difference alone makes it worth keeping in your toolbox even if you stay on Opus 4.8 for analytical work.
Where Fable 5 isn't the right choice
- Coding: still use Opus 4.8 or Sonnet. Fable 5 wasn't trained on code-heavy data — it'll produce code that compiles but isn't great.
- Math and reasoning: Opus 4.8 with extended thinking still wins clearly.
- Long-form factual content: Fable 5 is more willing to be evocative than precise. Avoid it for reports, briefings, or anything needing strict accuracy.
- Tool use and agentic flows: Opus 4.8 has better tool-call reliability for now.
Who is Fable 5 for?
- Novelists drafting fiction (especially anyone tired of GPT-5's sterile prose)
- Screenwriters and dialogue editors
- Game writers (NPC scripts, lore, branching dialogue)
- Marketing copywriters with strong brand voice
- Songwriters and poets
- Anyone with a creative writing practice who wants AI as a real collaborator, not a generic helper
Related reading
Bottom line
Fable 5 is the first frontier model that takes creative writing seriously as a category. If you write fiction, dialogue, or anything with brand voice, switch to it for those tasks today. Keep Opus 4.8 around for everything analytical. The era of "one model for everything" is over — Fable 5 is the proof.
