In the Text Arena on September 26, Claude Opus 5.5 with high Thinking level reaches the top position in creative writing. The still limited number of evaluations makes the result provisional.
Claude Opus 5.5 with high Thinking level is in first place in the Text Arena on September 26 for creative writing. The model also leads in the overall ranking. The results provide guidance but should still be read with caution in the writing category.
Top position with a narrow lead
In creative writing, Opus 5.5 scores 1521 points according to the Arena. Claude Fable 5 and Claude Opus 4.6 follow with 1501 points each. Overall in the Text Arena, Opus 5.5 leads with 1509 points just ahead of Opus 4.6 with 1505 points.
However, the first rank in the writing category is not yet securely established: there are 487 evaluations for Opus 5.5. The displayed rank range extends from first to seventh place. This means that the ranking position is associated with uncertainty given the existing evaluations.
Fewer em dashes in the evaluation
The Arena evaluation also compares the writing style of Claude models with high Reasoning level from August and September. Here, the em dash was counted significantly less in Opus 5.5 than in Opus 5: 0.8 instead of 15.2 times per 1000 words.
This metric describes a specific feature of the analyzed responses. However, it does not indicate which model is better suited for each writing task by itself. Also, the top position in an Arena ranking is not a guarantee that a model performs best across all text types or personal preferences.
What the result means for your model choice
When selecting an AI model for creative writing, you can consider Opus 5.5 based on this data. Take into account both the score and the small number of evaluations as well as the broad rank range. The gap to Opus 4.6 is 20 points in the writing category; overall it is four points.
For a practical decision, it is worth testing the models with the same writing tasks and comparing the responses based on your own criteria. The Arena ranking provides a reference point for this but does not replace looking at the texts you actually want to create.
Continue learning on TutKit
Sources
