GPT-Live can listen while it speaks.
To make that feel natural at ChatGPT scale, we rebuilt the voice stack from client to model.
This new architecture keeps audio flowing continuously, so deeper reasoning and tool use don't interrupt the conversation.
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.
Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.
We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security.
https://t.co/dKFCdpKd9v
We are committed to pushing the model frontier across cost efficiency, capability, and speed.
Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.
Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.
Anthropic 说他们为 Opus 5 删掉了 Claude Code 80% 的系统提示词。
我想验证下是不是。于是把 CLI 指向本地一个小服务器,抓它真正发出去的东西。
第一组数字就不太对。
Opus 4.7:15,225 字符。
Opus 4.8:4,467。
Opus 5:7,694。
他们是在 4.8 上删的,不是 5。Opus 5 的系统提示词比上一代大了 72%。
那 80% 确实是真的。每个版本里都同时带着两套 prompt,一个函数决定发哪套。判定条件是硬编码的子串匹配:model id 含 sonnet、haiku、claude-3-,或者等于 opus-4-0/4-1/4-5/4-6/4-7,走旧的。4.8 不在名单上。
只数手写的那几节策略正文,旧路径 12,443 字符,新路径 2,308。-81%。
删掉的是这几节:
# Doing tasks(3,321 字符)
# Executing actions with care(3,585)
# Text output(1,701)
# System(1,627)
# Using your tools(752)
# Tone and style(557)
换成一节 # Harness,1,702 字符,五条。这五条讲的全是环境长什么样,没有一条在规定该怎么做。markdown 会在终端里渲染,被拒绝的工具调用意味着用户说了不。
整套写注释的规矩,「默认不写注释」「绝不写多段 docstring」,只剩一句:
"Write code that reads like the surrounding code: match its comment density, naming, and idiom."
这个开关也支持用户手动控制。CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=1 能把短 prompt 套到任何模型上。我在 Opus 4.5 上试了一次,拿到的东西和 4.8 实际收到的逐字相同,只差写模型名和知识截止���期的那两行。
没料到的是 Opus 5 又开始往回加。
# Delivering work(2,019 字符)和 # Corrections(1,736),两段新内容,2.1.215 和 2.1.216 里完全没有。它们出现在 2.1.220,挂在一个 capability 上,而这个 capability 只有 claude-opus-5 有。4,467 加这两段,基本就是 Opus 5 的全部差值。
更巧的是 # Corrections 这一段。Anthropic 自己打包了一份 Opus 5 迁移指南,让第三方开发者往自己的 agent 里抄一段进去,抄的就是它,逐字相同。指南给的理由是 Opus 5 "flags and explains its own earlier mistakes at length, which reads as thrash in a user-facing product."
Fable 5 走的是第三条路,多一节 # Communicating with the user,8,911 字符。
所以「模型越强越不需要规则」这个说法不太站得住。删是上一代就删完的。换了新模型,翻车的方式也换了,规则就跟着��成一小块补丁,按模型贴。
如果你在这上面做东西,还有两件事。
model id 没被识别,走的又是 Bedrock、Vertex 或者自定义 base URL,就掉回那套 15K 的旧 prompt。它查的是你的 provider,不只是模型。
同一个开关也在分流工具描述。53,328 字符降到 37,167。你的上下文大头根本不在 system prompt 里。
想自己验?可以写个小 HTTP 服务,把 response 存下来,直接返 400。把 ANTHROPIC_BASE_URL 指过去,然后 claude -p "hi" --model 模型id。
Anthropic 删自己的规则我没什么感觉。难受的是我也写过一堆同样的规则,其中不少大概只是在替模型兜底,而它早就不需要人兜了。哪些是,我分不出来,只能一条条删掉看会不会坏。
@trq212 的原帖值得读全文。
Claude Code now has a screen reader mode.
Running `claude --ax-screen-reader` swaps the visual terminal UI for plain, linear text that screen readers (like VoiceOver and NVDA) can follow.