Is Claude Opus 5 an improvement for writers?
- Reading time
- 3 min read
- Written by
- by Kate
This week, I sat down on stream to work out my thoughts on Opus 5. My relationship with the Claude 5 series has been… complicated. Fable showed a remarkable ability for planning and analysing stories, but was far too expensive to recommend as a writing partner. Sonnet 5 was fine, but didn’t feel like an improvement on the 4 series for writing.
So the question going into Opus 5 was whether the model would pull the series together for writers, or whether we’d be left with another tradeoff?
A Great Brainstorming Buddy
Don’t let my lack of overt praise deter you. Opus 5 has its strengths. The model is great for brainstorming and responds well to prompting. Hand it an outline, and it’ll find connections you missed and poke holes in your story logic. If you give it a summary and ask “what doesn’t work here?”, it’ll give you a thoughtful answer.
Towards the end of the stream, I spent time with Opus 5 looking into a Scottish folklore-inspired romance, and it delivered. Together, we created a charming but tragic story about a glaistig and a descendant of the family she protects—a romance built around change, and how traditions and stories fade and die. It found thematic threads quickly and built on them without needing too much guidance.
Every Sentence Needs a Metaphor
Like its Sonnet and Fable siblings, Opus heavily leans towards metaphorical language. If a simile or metaphor could be used to describe a situation, it will be there. Opus 5 seems to really rely on these, layering up the flowery language to make a metaphor sandwich. If grief is present, we are hammered over the head to make sure we know it.
The box coughed. A crack went through the metal, hard as a dropped spanner on concrete, and blue light spat out past my wrist and licked the wall.
On top of that, there’s a habit of over-explaining. Where the natural flow of a scene might call for a character’s silence, Opus 5 tends to narrate the emotion explicitly and then summarize what just happened. It announces realizations rather than letting the reader piece them together. If you’ve ever received feedback that says “trust your reader more,” Opus 5 could use the same note.
Has a Short Attention Span
As AI models get more capable and more “agentic,” writing quality doesn’t always follow. The biggest issue I ran into was a decay in quality; the model follows style and instructions well for the first few hundred words, but the longer the output goes, the more it drifts.
This can be countered by working in shorter beats rather than asking for full scenes in one generation, but it means more back-and-forth and more tokens spent. It is not inherently bad, but it did make me feel like I was writing in 2024 again.
I also found that Opus 5 has a very strong inbuilt voice. Even after creating a new style guide specifically for Opus, by the end of the prose generation, bad habits were reappearing in the prose.
Opus 5 also has a tendency to repeat input writing. Hand it a character description, or some prose before the beat, and your words show up nearly verbatim in the output. Whilst it doesn’t occur in every generation, it was frequent enough to be frustrating, and the repetition feels clumsy. In the example below, the repetition occurred 117 words after the original mention.
I grinned at the junction box so she couldn’t see it. Twisted the cores together, like plaiting hair, the way Ma’s ledger showed in diagrams made by machines long gone. Copper doesn’t care if you’re cold. I appreciated the honesty. I tucked the cords back into place.
I got the cover plate up with the heel of my hand and shoved, trying to seat it, trying to get the splice buried and insulated before whatever was coming came. The plate caught on a lip of rust. The whine went up another note. I could feel it in my fillings.
Copper doesn’t care if you’re cold. Doesn’t care if you’re quick, either.
Personally, if I want to use an Opus model for prose generation, I will be sticking with Opus 4.6. Opus 5 has its places, but the prose quality has decreased, and the new tokenizer has increased the cost of otherwise identical prompts, whilst giving poorer quality responses.
If you want to interrogate an outline, or use it as a brainstorming partner, it shows much more promise. It’s up to you to decide if the price is worth it.
Kate
Based in the UK, Kate has been writing since she was young, driven by a burning need to get the vivid tales in her head down on paper… or the computer screen.