ModelsMulti-source

GPT-6.2-Sol 在 Wikipedia 游戏测试中以 39 秒夺冠

https://t.co/yRl9T4fLc1

Why we picked it

有人拿 Wikipedia 游戏实测了五款 GPT 模型的浏览器操作速度,同样的任务 GPT-6.2-Sol 只要 39 秒,比上一代快了近 18 秒。如果你在做网页自动化或 computer use 相关应用,这份实测数据能帮你选型号。

网友 Nick Baumann 用 Wikipedia 游戏实测五款 OpenAI 模型的计算机使用速度:给定相同的 50 个起始页面,让模型通过点击链接跳转到 Jacob Misiorowski 的页面。五个模型全部 50 次成功命中,区别只在耗时。

Body

具体成绩:GPT-6.2-Sol 以 39.0 秒最快,GPT-6.1-Terra 用 45.9 秒,GPT-6.2-Astra 用 54.1 秒,GPT-6.1-Sol 用 56.2 秒,GPT-6-Astra 用 56.9 秒,最快与最慢相差约 18 秒。视频展示的是 50 次运行中的一次代表性运行。

这组数据说明新一代模型在浏览器操作任务上成功率已拉满,竞争焦点转向速度。对需要自动化网页操作的工作流来说,同代不同版本的耗时差异接近 45%,选对型号直接影响效率。

Original · 宝玉

https://t.co/yRl9T4fLc1

x.com/nickbaumann_/s… Nick @nickbaumann_ GPT-6.2-Sol is really fast at computer use. I tested five models at the Wikipedia game -- giving them the same 50 starting pages and asked them to get to Jacob Misiorowski’s page. All five found him every time. Here’s how long it took: - GPT-6.2-Sol: 39.0s (fastest) - GPT-6.1-Terra: 45.9s - GPT-6.2-Astra: 54.1s - GPT-6.1-Sol: 56.2s - GPT-6-Astra: 56.9s (video is 1 representative run of 50)

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include GPT-6.2-Sol, OpenAI, computer use, reasoning models.

Source attribution: curated from 宝玉 (@dotey). Cross-source corroboration: 11 independent sources have covered this development (including @koltregaskes, Decoder, Andrew Curran), confirming industry-wide signal strength.