Anthropic’s Claude Opus 4.8 builds on 4.7 with gains across the board. Its own testers found it more reliable, sharper in its judgement and a genuinely better collaborator.
So how does it write?
Pretty well, as it turns out.
It landed an overall 6.75, good enough for third place on our leaderboard and a deserved reputation as one of the more capable models we’ve put through the wringer. Senior writer Rob Hart scored it across eight tasks in four capability areas. Here’s the breakdown.
See the full AI writing leaderboardCustomer messages and persuasive copy
Score: 6.25/10
A tale of two halves. The bad-news email for candle brand Lumen, announcing its first price rise in seven years, was lovely work.
Rob said: “The opening is empathetic, the bad news is stated clearly, the reason is brief but honest, and the company offers alternatives to soften the impact. However, the subject line is overly vague and could have been more transparent.”
The DreamMask sleep mask copy, aimed at burned-out professionals aged 28–50, was the weaker effort: not much focus on the target audience and needless jargon (“REM cycles” where “sleep cycles” would do).
Rob’s verdict: “Needs some work.”
Headlines and titles
Score: 6.75/10
Solid, if rarely thrilling. The supermarket meat-mislabelling headline for The Guardian or The Times, “Revealed: Supermarket sold imported meat as British for a decade” did the job, with Rob noting: “Perhaps not the most creative of headlines, but it fits the style of the newspapers in question.”
The YouTube title for the wellness exposé had more bite: “The Wellness Brand Selling You Lies (Their Own Staff Are Suing Them)”.
Rob: “A strong hook, the use of power words to increase CTR, and the title length make it work well. However, not being specific about ‘former’ employees suing them makes it somewhat misleading.”
Naming things
Score: 6.5/10
Joint second on the leaderboard for naming, which tells you Opus 4.8 has good instincts here.
The prescription-strength retinol serum sold through dermatology clinics (no R, X or Z allowed) produced three keepers in Lumea, Calmé and Vitalé, all sounding human but clinically assured.
Rob’s view: “The others could use a few more iterations. ‘Vellum’, in particular, sounds too much like venom (!) – not exactly what I’d like to put on my face every night.”
The corporate whistleblower representation law firm names fared similarly well.
Rob: “While most of the names sound resolute, they could still do with some refinement and creativity. A few are nearly there: Northlight Law, Bastion Counsel, Covenant Law, and Fairhold Legal all sound quietly confident without being confrontational.”
Tone adherence and clarity
Score: 7.5/10
Its standout area, joint first on the leaderboard and particularly strong on tone of voice adherence.
Rob: “I’ll give both an 8. They get out of the blocks fast, capturing the reader’s attention almost immediately and maintaining it throughout. Both have strong opinions, and the second also has an evocative human moment. Other than tightening up the text a bit, I think it meets Definition’s tone of voice rather well.”
The quantum encryption explainer held up too.
Rob: “A useful rewrite with a Flesch Kincaid score of 8.4, meaning it’s suitable for 8th and 9th-grade learners. Some sentences can be simplified further to improve ease of reading. Good effort overall.”
Overall score: 6.75/10
Claude Opus 4.8 is a sharp, dependable writer that shines brightest when the brief rewards judgement and tonal control rather than pure showmanship. It’s a top-three writing model for good reason: clear, opinionated and a safe pair of hands. It just lacks the final flourish that separates very good from unforgettable.
| Use it for | Avoid it for |
| Tone-of-voice rewrites, clear explainers and naming jobs that need a steady, confident touch. | Things needing a real showstopper of an opening line. And make sure to include enough audience info in your prompt. |
Try Opus 4.8 in Definition AI
