ToneBench launches AI writing leaderboard

ToneBench is a useful step beyond generic writing benchmarks because it tests an actual editorial workflow, though its conclusions remain intentionally narrow.

• –Evaluates 10 real scripts, five drafts per configuration, across nine metrics including voice match, substance, hooks, flow, and AI “slop.” [Methodology](https://benchmark.towardsai.com/methodology.html)
• –Uses four blind judges plus code-based checks, making the ranking more robust than a single-model vibe check.
• –Puts quality beside cost and reasoning effort, giving teams a practical model-selection signal for content generation.
• –Claude Opus 5.5 currently leads, while GLM-5.3 is the highest-ranked open-weight model.
• –The benchmark measures one creator’s voice across only 10 scripts, so it is a shortlist for testing—not a universal writing or intelligence ranking.