Anthropic's Claude Opus 4.8 achieved a score of 0.925 on the Editorial Craft benchmark, ranking seventh among 12 leading AI models tested on publication-ready writing tasks, while OpenAI's GPT-5.6 Sol Pro dominated the evaluation with a 0.97 score and claimed four of the top six positions overall. The benchmark tested each model on 12 real-world editorial assignments without external tools or retries, measuring their ability to draft articles from reporting notes, rewrite copy for clarity, reconstruct narratives from scrambled information, and weave quotations naturally into stories.

Two tasks in the Editorial Craft battery were graded against ground truth answers, while 10 open-ended writing assignments were evaluated blind by GPT-5.4 against a fixed rubric. All models had their reasoning effort set to maximum where available, ensuring a fair comparison across different architectures. The benchmark's focus on editorial judgment rather than abstract reasoning reflects the kinds of tasks that newsrooms and content teams actually perform.

How Do Claude's Models Stack Up Against the Competition?

Anthropic placed three models in the top 10, with Claude Opus 4.8 leading the company's performance. Claude Sonnet 5 scored 0.883, placing ninth overall, while Claude Fable 5 scored 0.800 and ranked tenth. The spread between Anthropic's flagship and mid-tier models reflects different design priorities: Opus targets maximum capability, Sonnet balances performance with cost efficiency, and Fable emphasizes affordability.

• Claude Opus 4.8: Scored 0.925 with an estimated cost of $0.0160 per task, ranking seventh overall and representing Anthropic's strongest editorial performance
• Claude Sonnet 5: Scored 0.883 with an estimated cost of $0.0063 per task, demonstrating solid mid-range capability at lower cost
• Claude Fable 5: Scored 0.800 with an estimated cost of $0.03 per task, the lowest-scoring Anthropic model but designed for budget-conscious deployments

The benchmark reveals a critical trade-off in the AI market between performance and cost. While Claude Opus 4.8 achieved competitive accuracy, its estimated cost of $0.0160 per task was roughly 2.5 times higher than Claude Sonnet 5 and significantly more expensive than OpenAI's Luna models, which cost $0.0007 per task despite scoring 0.942 and 0.933 respectively.

SpaceXAI's Grok 4.6 scored 0.933 at rank five, positioning it as the leading non-OpenAI model ahead of Claude Opus 4.8. Google's Gemini 3.7 Flash ranked eleventh with a score of 0.775, while Phi-4-reasoning scored 0.475.

What Does This Benchmark Actually Measure?

The Editorial Craft evaluation specifically targets tasks that reveal whether AI can meaningfully assist human editors. The 12-task battery includes drafting publication-ready articles from reporting notes, rewriting copy for clarity and narrative flow, reconstructing chronological narratives from scrambled information, and weaving quotations naturally into stories. These are not abstract reasoning challenges but practical writing problems that determine whether AI can catch common editorial failures: mechanical transitions, promotional language, poor sentence rhythm, and attribution errors.

OpenAI's dominance across the leaderboard suggests that the company's recent model releases have achieved a measurable edge in tasks requiring nuanced language judgment and editorial sensibility. GPT-5.6 Sol Pro's 0.97 score represents near-perfect performance on a battery designed to expose weaknesses in AI writing that human editors would immediately flag.

How to Use These Results When Choosing an AI Writing Tool

• Evaluate Your Budget: If cost per task is critical, OpenAI's Luna models offer strong performance at $0.0007 per task, while Claude Sonnet 5 provides a middle ground at $0.0063 per task
• Test on Your Workflows: The full Editorial Craft leaderboard includes interactive access to every prompt, answer, and per-task grade, allowing teams to evaluate which models perform best on their specific editorial tasks rather than relying on aggregate scores alone
• Consider Your Use Case: For maximum editorial quality, Claude Opus 4.8 and OpenAI's top-tier models offer the highest scores, but mid-tier options like Claude Sonnet 5 may deliver sufficient quality at significantly lower cost for many newsrooms

For organizations evaluating AI writing assistants, the benchmark highlights a widening performance gap between leading models and mid-tier alternatives. Claude Opus 4.8's seventh-place finish demonstrates that Anthropic remains competitive at the highest tier, but the cost differential and OpenAI's consistent placement across multiple model variants suggest that editorial teams may face pressure to adopt OpenAI's products if performance and budget constraints align.

The benchmark also signals broader shifts in the AI market. Anthropic's inclusion of Claude Fable 5 in the evaluation reflects growing demand for lower-cost models, even if they sacrifice some performance. The presence of Grok 4.6 ahead of Claude Opus 4.8 indicates that SpaceXAI has made meaningful progress in language quality, challenging Anthropic's historical dominance in writing-focused tasks.