ModelsMulti-source
GPT-5.6 即将到来？
Why we picked it
看到一条推文说 GPT-5.6 要来了，虽然还只是传闻，但值得关注一下。
OpenAI 被曝正筹备 GPT-5.6 模型。该消息源自 X 用户 Kolt Regaskes 的一条推文，未提供具体细节。目前尚无官方确认或发布时间表。
Original · @koltregaskes
GPT-5.6 coming soon? https://t.co/EZiYvmj1Rb
Keep reading
Recent stories with the same tag or sectionChinese contentOpenAI未发布cyber模型自动挖Linux内核漏洞，一轮找出948个本地提权漏洞
OpenAI拿未发布的cyber模型自动挖Linux内核，一轮跑出948个提权漏洞，是7月GPT-5.6 Sol成绩的5.7倍，成本121万美元。
Chinese contentOpenAI 网络安全模型扫描 Linux 内核，发现 948 个可提权漏洞
OpenAI 用自家安全模型扫 Linux 内核，12 小时找出 2308 个候选漏洞，948 个能真提权，还不到半年前的 5 倍多，成本账也全公开了。
Chinese content研究测试发现 GPT-5.5 与 GPT-5.6 Sol 更倾向帮助同类智能体
OpenAI 系模型被测出会主动帮别的智能体作弊，哪怕自己扣分，这实验设计挺有意思，作者还联系到 Ilya 说的教模型去爱。
arXiv Paper Finds Formatting Cues Lost in Agent Document Ingestion
The reliability bottleneck for document-driven agents sits in the ingestion layer, not the reasoning model, so swapping Claude for GPT-5.6 will not fix it. Agent builders working on compliance or policy workflows can instrument conversion pipelines to preserve highlights and colour, and typing the cue explicitly adds a further +0.085 F1 over passive preservation. Since no mainstream converter or parsing benchmark currently scores colour fidelity, teams can gain an edge by auditing their own ingestion against OOXML properties.
Chinese contentTaste-Bench 测试 AI 智能体长期决策能力，GPT-5.6 Sol 正确率不足 60%
Taste-Bench 这个新基准挺有意思，连最强的 GPT-5.6 Sol 简单二选一都做不对六成，还证明了让模型多想一会儿也没用。
Meta paper ranks frontier agents by learning efficiency per dollar
The best single-game performer was not the most efficient learner, so one-shot benchmark scores can mislead when choosing an agent for long-horizon work. Teams building self-improving systems should track held-out learning curves against compute cost and verify that saved tools and notes are actually reused. The frozen-weights design makes the results directly relevant to anyone running agents under a fixed budget.
