要点
What matters
10 月 16 日复查 Arena Single Image Edit 页面，页面仍标注 10 月 6 日：59 个模型、30,985,932 票、分数范围 998–1524。前两行仍为 gpt-image-2.5-sunburst（1524±5、59,825 票、Preliminary）和 gpt-image-2.5-flare（1481±5、57,408 票、Preliminary）。
On October 16, Arena’s Single Image Edit page remains dated October 6: 59 models, 30,985,932 votes and a 998–1524 score range. The first two rows remain gpt-image-2.5-sunburst (1524±5, 59,825 votes, Preliminary) and gpt-image-2.5-flare (1481±5, 57,408 votes, Preliminary).
范围与事实边界
Scope and limits
这与 10 月 15 日保存的可比快照一致，因此不声称排名、票数或产品能力变化。Arena 是偏好榜；Preliminary、分数区间和票数都不证明产品上线、账户可用、商用权利或所有商品图任务的可交付性。
This matches the comparable October 15 snapshot, so no ranking, vote-count or capability change is claimed. Arena is a preference leaderboard; Preliminary labels, score intervals and votes do not establish product release, account access, commercial rights or deliverability for all product-image tasks.
编辑工作台
Editorial workbench
只从榜单建立候选池。用已获授权素材按文字、身份、透明边缘、局部编辑和导出规格评分，同时记录计划资格、成本、许可和失败原因；未通过不能进入生产。产出是一份模型候选验收表。
Use the leaderboard only to make a candidate pool. Score authorized assets on text, identity, transparent edges, local edits and export specs while recording plan eligibility, cost, license and failure reasons; failures do not enter production. The deliverable is a model-candidate acceptance sheet.