不知道是快讯还是慢讯，grok4.3beta已经放出来了。 3 个帖子 - 3 位参与者 阅读完整话题
今天打开tg，突然跳出来让选择语言，看到有简体中文，设置成简体中文后，和之前用的第三方还是有一些不一样的，大家赶紧去体验吧 2 个帖子 - 2 位参与者 阅读完整话题
如题。虽然好像没什么用 4 个帖子 - 4 位参与者 阅读完整话题
2 个帖子 - 2 位参与者 阅读完整话题
之前听说chatgpt要开始推送广告了，现在第一次发现真推送了 5 个帖子 - 5 位参与者 阅读完整话题
2026-4-14-22:53 对不起大家，我火星了，不是故意水一贴的 不知道是我火星了，还是难得快讯一次 12 个帖子 - 7 位参与者 阅读完整话题
We’ve identified that intermittent 401 errors on some API endpoints are being caused by timeouts. We have applied mitigations that reduced impact in parts of the system and are continuing to deploy additional fixes and monitor recovery until error rates return to expected levels. 1 个帖子 - 1 位参与者 阅读完整话题
github.com/anthropics/claude-code [BUG] Disabling telemetry also disables 1-hour prompt cache TTL 已打开 07:02PM - 08 Apr 26 UTC EmpireJones bug has repro platform:windows area:core api:anthropic ### Preflight Checklist - [x] I have searched [existing issues](https://github. … com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) and this hasn't been reported yet - [x] This is a single bug report (please file separate reports for different bugs) - [x] I am using the latest version of Claude Code ### What's Wrong? When telemetry is disabled (e.g., via DISABLE_TELEMETRY=1 or CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1), sessions that would otherwise receive 1-hour prompt cache TTL appear to fall back to the 5-minute TTL instead. ### What Should Happen? Prompt cache TTL selection should be independent of telemetry. If a user/session would otherwise qualify for 1-hour prompt cache TTL, it should still receive it even when telemetry is disabled. ### Error Messages/Logs ```shell ``` ### Steps to Reproduce 1. Start a session with telemetry enabled. 2. Confirm the session is using the 1-hour prompt cache tier. 3. Start a comparable session with either `DISABLE_TELEMETRY=1` or `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1`. 4. Compare the cache metadata between the two sessions. 5. Observe that the 1-hour TTL is no longer applied when telemetry is disabled. Note: This may not be reproducible if using bedrock and `ENABLE_PROMPT_CACHING_1H_BEDROCK=1` is set. I only tested with an Anthropic Claude Max plan. What metadata to check: In session transcripts, check the assistant response `usage.cache_creation` metadata for the prompt cache tier: - `usage.cache_creation.ephemeral_1h_input_tokens` - `usage.cache_creation.ephemeral_5m_input_tokens` If 1-hour TTL is active, `ephemeral_1h_input_tokens` should be non-zero. If 5-minute TTL is active, `ephemeral_5m_input_tokens` should be non-zero. ### Claude Model Opus ### Is this a regression? I don't know ### Last Working Version _No response_ ### Claude Code Version 2.1.96 ### Platform Anthropic API ### Operating System Windows ### Terminal/Shell VS Code integrated terminal ### Additional Information _No response_ 让 cc 分析之前泄露的源码原因如下，看起来是个 bug： 8 个帖子 - 4 位参与者 阅读完整话题
有没有和我一样最近没关注这个等着4月底再搞一个的佬友 这波被谷歌阴了，没想到还能回退的 6 个帖子 - 4 位参与者 阅读完整话题
以前是每天2刀额度，但快手自家模型免费用，现在没有免费的快手模型用了，新注册就给一个月有效期的50赠金 如果是以前注册的，不会给你刷这50赠金，老用户不如狗这波，我就用了2刀 模型好像比以前多了gpt，但倍率有点高 看了下公告是4月2号的信息 1 个帖子 - 1 位参与者 阅读完整话题