Arena · 2026-10-19 · REVIEWED 2026.10.24

Arena 单图编辑页仍为 60 模型、3,140 万票:停用模型在榜不等于仍可用

Arena single-image-edit remains at 60 models and 31.4M votes: a retired model on a board is not availability

固定观察 · 单图编辑偏好榜

Fixed observation / single-image-edit preference leaderboard

Jingwen He

要点

What matters

10 月 24 日打开 Arena Single Image Edit 页面时,页面日期仍为 10 月 19 日:60 个模型、31,404,772 票、分数区间 996–1515。前两行仍为 GPT-Image-2.5 Sunburst(1514±6、82,078 票、Preliminary)和 Flare(1481±5、79,506 票、Preliminary)。页面第 56 行仍列 GPT-Image-1(1139±3、2,869,690 票),但这不与 OpenAI 的 10 月 23 日停用状态矛盾:榜单保留的是历史比较记录,而非账户可用性目录。

When opened on October 24, Arena’s Single Image Edit page remained dated October 19: 60 models, 31,404,772 votes and a 996–1515 score range. The first two rows remain GPT-Image-2.5 Sunburst (1514±6, 82,078 votes, Preliminary) and Flare (1481±5, 79,506 votes, Preliminary). Row 56 still lists GPT-Image-1 (1139±3, 2,869,690 votes), which does not contradict OpenAI’s October 23 retirement: the board preserves historical comparisons rather than serving as an account-availability catalogue.

范围与事实边界

Scope and limits

This is a dated preference-leaderboard observation, not evidence of a launch, access, pricing, licensing, rights or universal production quality. Scores, uncertainty intervals, vote counts and Preliminary labels do not prove text rendering, product/person identity, transparent edges, regional suitability or an ability to migrate a production flow. Compared with the saved October 23 snapshot, no comparable numeric movement was observed.

This is a dated preference-leaderboard observation, not evidence of a launch, access, pricing, licensing, rights or universal production quality. Scores, uncertainty intervals, vote counts and Preliminary labels do not prove text rendering, product/person identity, transparent edges, regional suitability or an ability to migrate a production flow. Compared with the saved October 23 snapshot, no comparable numeric movement was observed.

编辑工作台

Editorial workbench

将榜单只用于形成候选名,并单列“当前可用”证据。每个候选都用授权素材测试文字、身份、边缘、局部编辑、导出、地区/套餐与失败回退;把排行榜快照日期和账户检测日期分开记录。产出是一张候选—可用性双轨表。

Use the board only to form candidate names, and record separate evidence of current availability. Test each candidate with authorized assets for text, identity, edges, local editing, export, region/plan and failure fallback; record the leaderboard snapshot date separately from the account-check date. The deliverable is a candidate-versus-availability dual-track sheet.

核验来源

Sources checked

← 返回今日← Back to today