Claude vs ChatGPT for writing: what 42 scored drafts show
Claude wrote better blog posts than ChatGPT's GPT models in our test. Claude Opus 5.5 scored 93.8 out of 100 and Claude Sonnet 5.5 scored 92.8. The best GPT model, GPT 5.4 Mini, scored 90.2. For copywriting it's close to a tie, and GPT 5.4 Mini is the cheapest good writer of the seven at 0.9 cents a post.
Written October 16, 2026 by Preston Vawdrey at Draftly.
Every number here comes from our 14-model writing benchmark, run October 6, 2026. Draftly offers both families, so either result works for me.
Blog posts: Claude's newest models lead
Each model wrote three blog posts for three made-up businesses, under the same writing rules. A script scored formatting, two AI judges scored how human the writing sounds, and the overall score averages the two.
Claude Opus 5.5 ranked 3rd of all 14 models and Claude Sonnet 5.5 ranked 4th. GPT 5.4 Mini came 5th. The two bigger GPT models finished 7th and 10th.
The version matters more than the brand. Claude Sonnet 4.6 scored 81.5, 11.3 points below Sonnet 5.5, and Claude Haiku 4.5 came last of the 14 at 74. If you write with an older Claude model, GPT 5.4 Mini beats it.
| Model | App | Overall | Formatting | Voice | SEO checks (of 18) | Cost per post | Seconds | Avg words |
|---|---|---|---|---|---|---|---|---|
| Claude Opus 5.5 | Claude | 93.8 | 95 | 92 | 18 | $0.099 | 50 | 1,214 |
| Claude Sonnet 5.5 | Claude | 92.8 | 91 | 95 | 17 | $0.049 | 37 | 1,008 |
| GPT 5.4 Mini | ChatGPT | 90.2 | 94 | 87 | 15 | $0.0092 | 13 | 1,275 |
| GPT 5.5 | ChatGPT | 87.6 | 85 | 90 | 18 | $0.113 | 51 | 1,774 |
| GPT 5.4 | ChatGPT | 81.8 | 85 | 79 | 13 | $0.034 | 40 | 1,429 |
| Claude Sonnet 4.6 | Claude | 81.5 | 79 | 84 | 13 | $0.027 | 38 | 924 |
| Claude Haiku 4.5 | Claude | 74 | 68 | 80 | 10 | $0.010 | 20 | 1,098 |
Where each one sounds like AI
GPT 5.5 and GPT 5.4 each used 5contrast phrases ("it's not X, it's Y") across three posts, against 2 for each of the Claude 5.5 models. GPT 5.5 also ran long: it averaged 1,774 words on a 1,000 to 1,200 word brief, and none of its three posts landed in range.
The older Claude models have the worst em dash habit we saw. Claude Haiku 4.5 used 29 of them in three posts and Sonnet 4.6 used 9, even with a rule telling them not to. Both Claude 5.5 models used none.
| Model | App | Em dashes | Contrast phrases | AI vocabulary | Posts in word range |
|---|---|---|---|---|---|
| Claude Opus 5.5 | Claude | 0 | 2 | 1 | 3 of 3 |
| Claude Sonnet 5.5 | Claude | 0 | 2 | 1 | 1 of 3 |
| GPT 5.4 Mini | ChatGPT | 3 | 2 | 0 | 3 of 3 |
| GPT 5.5 | ChatGPT | 0 | 5 | 1 | 0 of 3 |
| GPT 5.4 | ChatGPT | 0 | 5 | 0 | 3 of 3 |
| Claude Sonnet 4.6 | Claude | 9 | 6 | 0 | 1 of 3 |
| Claude Haiku 4.5 | Claude | 29 | 2 | 3 | 3 of 3 |
Which sounds more human
Claude Sonnet 5.5 had the highest voice score of all 14 models, at 95. Claude Opus 5.5 scored 92. The best GPT voice score was GPT 5.5's 90, and GPT 5.4 Mini scored 87.
One of the two judges was GPT 5.4, an OpenAI model, which could tilt things toward GPT. It went the other way: each judge scored its own model lower than the other judge did. GPT 5.4 gave itself 77, and the Gemini judge gave it 81.
Copywriting: close to a tie
Each model also wrote a five-part copy pack per business, including Google Ads headlines and descriptions. Claude Opus 5.5 scored 90.8, Claude Sonnet 5.5 89.6 and GPT 5.5 88.8. With three packs per model, a gap that small is a tie.
The real difference is character limits. Every GPT model and both Claude 5.5 models met all 15 Google Ads limits. Claude Sonnet 4.6 met 8 and Claude Haiku 4.5 met 7, so their ads would need trimming before you could upload them.
| Model | App | Overall | Clarity | Persuasion | Voice | Specificity | Ad limits met | Cost per pack |
|---|---|---|---|---|---|---|---|---|
| Claude Opus 5.5 | Claude | 90.8 | 8.5 | 7.2 | 8.8 | 8.2 | 15 of 15 | $0.053 |
| Claude Sonnet 5.5 | Claude | 89.6 | 8.5 | 6.8 | 8.5 | 7.8 | 15 of 15 | $0.030 |
| GPT 5.5 | ChatGPT | 88.8 | 8.5 | 6.5 | 8.3 | 7.7 | 15 of 15 | $0.026 |
| GPT 5.4 Mini | ChatGPT | 87.6 | 8.5 | 6.5 | 8 | 7.5 | 15 of 15 | $0.0017 |
| GPT 5.4 | ChatGPT | 87.1 | 8.5 | 6.2 | 7.8 | 7.2 | 15 of 15 | $0.0054 |
| Claude Sonnet 4.6 | Claude | 85.5 | 8.5 | 7.2 | 8.3 | 8 | 8 of 15 | $0.0077 |
| Claude Haiku 4.5 | Claude | 80.5 | 8.3 | 6.5 | 7.5 | 7.2 | 7 of 15 | $0.0025 |
Cost and speed
GPT 5.4 Mini is the bargain. It cost 0.9 cents a post and finished in 13 seconds, the fastest of all 14. Claude Sonnet 5.5 cost 4.9 cents and took 37 seconds.
Paying more didn't buy a better post on the GPT side. GPT 5.5 cost 11.3 cents a post, the most of these seven, and scored 2.6 points below GPT 5.4 Mini. These are API prices from our test, which is what a tool like Draftly pays. A ChatGPT or Claude subscription is a flat monthly fee.
Which to pick for each job
Everyday blog posts
Claude Sonnet 5.5
Highest voice score of all 14 models (95) at 4.9 cents a post.
The best post, cost aside
Claude Opus 5.5
93.8 overall, the top Claude or GPT score, and all 18 SEO checks passed.
Lots of posts on a small budget
GPT 5.4 Mini
90.2 overall for 0.9 cents a post, done in 13 seconds.
Ads and short copy
Claude Opus 5.5
Top copy score here (90.8) and every Google Ads limit met. Any GPT model also hit all 15.
If you already pay for one app, stay there and switch models inside it. The version you pick moved the score more than the company did.
What this test can't tell you
- It tested models, not apps.We called each model through the Vercel AI Gateway with Draftly's writing rules. ChatGPT and Claude add their own instructions and pick models for you.
- Three briefs per model is a small sample. Treat a gap of two or three points as a tie.
- Models change fast.This is Edition 1. OpenAI's GPT 6 models, including GPT 6 Luna, weren't in it. The next edition adds them and more briefs.
- It doesn't check facts or sales. It scores rule-following and how human the writing sounds, not whether the facts are right or whether readers buy.
The full method, all 14 models and real excerpts are on the benchmark page.
Write with either one in Draftly
Draftly writes with Claude Sonnet 5.5 by default. Paid plans can switch any post to Claude Opus 5.5, Claude Haiku 4.5, Claude Sonnet 4.6, GPT 5.5, GPT 5.4, GPT 5.4 Mini or GPT 6 Luna, and every model gets the same writing rules and the same check for AI-sounding writing. Credits per post are on the pricing page.
If you'd rather keep writing in the Claude or ChatGPT app, the Draftly MCP connector brings your brand voice and past posts into the chat.
Keep reading
BenchmarkBest AI for writing: 14 models testedWhich models write the most human-sounding posts, and what each one costs.
GuideSEO copywriting rules we check on every postStructure, voice and call-to-action rules, with before and after examples.
Blog postAEO and GEO for small business blogs in 2026How to get your small business blog cited by ChatGPT, Claude and Google AI Overviews, based on Google's own guidance…
Frequently asked questions
Is Claude better than ChatGPT for writing?
For blog posts, yes, in our test. Claude Opus 5.5 scored 93.8 and Claude Sonnet 5.5 92.8, against 90.2 for the best GPT model, GPT 5.4 Mini. For copywriting the two were close: Claude Opus 5.5 scored 90.8 and GPT 5.5 88.8.
Which Claude model is best for writing?
Claude Opus 5.5 had the higher overall score (93.8). Claude Sonnet 5.5 had the best voice score of the 14 models we tested (95) at about half the cost per post, so it's the one we'd use day to day.
Which ChatGPT model is best for writing?
GPT 5.4 Mini, at 90.2 overall. GPT 5.5 scored 87.6 and GPT 5.4 81.8. The bigger models lost points for running long and for contrast phrasing.
Is Claude or ChatGPT cheaper for writing blog posts?
GPT 5.4 Mini is the cheapest, at 0.9 cents a post through the API. Claude Sonnet 5.5 cost 4.9 cents and Claude Opus 5.5 9.9 cents. GPT 5.5 was the most expensive of the seven at 11.3 cents.
Did you test the ChatGPT and Claude apps?
No. We called each model through the Vercel AI Gateway with the same writing rules, so the scores show how each model follows instructions. The chat apps add their own instructions and choose models for you, so your results there can differ.
Can I write with Claude and ChatGPT models in Draftly?
Yes. Draftly writes with Claude Sonnet 5.5 by default, 150 credits a post. Paid plans can switch any post to Claude Opus 5.5, GPT 5.5 or GPT 5.4 for 250 credits, GPT 5.4 Mini or Claude Haiku 4.5 for 150, or GPT 6 Luna for 50.