Unverified45% confidenceFactTime unknown

GPT-5.4在Online Mind2Web上仅靠截图观察力拿下92.8%，而ChatGPT Atlas智能体模式只有70.9%

1

Sources

45%

Confidence

Medium-term (~90 days)

Relevance

5/31/2026

First Seen

Valid until: 8/29/2026

Sources

GPT-5.4深度评测：原生计算机使用、推理编程合体，OpenAI重回王座

bilibiliTerminator-AI

Related Entities

GPT-5.4model ChatGPT Atlasproduct Mind2Webconcept

Related Claims

UnverifiedGPT-5.6编程能力在关键基准上达到91.9%，安全评测CyberGym得分85.6%，均超越竞品Nexus 573% similar Unverified2023年研究者在 GPT-2 中识别出负责间接宾语识别的完整电路，涉及不足1%的注意力头73% similar UnverifiedGPT-5.6 Sol在BrowseComp浏览任务基准上达到92.2%73% similar Unverified谷歌DeepMind的FactScore显示GPT-4在人物传记生成任务中精确事实陈述比例约为73%72% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/13754

API

curl https://kongchang.com/api/v1/knowledge/claims/13754

MCP

get_claim(id=13754)