当前速蹬窗口 官方预告后,到额度重置前
等待官方预告,持续监控中。
监控 Codex 重置窗口、预测重置概率,并追踪模型智商状态。
等待官方预告,持续监控中。
低概率:刚完成补偿重置,短期进入冷却。
大模型与小加一公众号
Codex 雷达蹬友群
只在官方速蹬窗口开启时发邮件;预测信号不会触发提醒。
按北京时间每日最后一次预测聚合,读的是未来 24-48 小时趋势;窗口开启当天概率偏低是正常的,括号内是该窗口持续时长,0分钟表示即时重置。
6月4日:6/12 通过,基线 8/12
开发中的实验性功能:使用固定 DeepSWE 12 题探针的通过率折算为 IQ 指数趋势;样本、口径和自动化仍在校准中。
过去 24 小时观测到 128 条公开动态,其中 66 条进入候选信号;本轮新增 61 条,已见过 5 条。预测模型只处理新增候选。
Increased latency for Codex compaction for a subset of users
What happened when one of our models found a counterexample to an 80-year-old Erdős conjecture? Researchers @alexwei_, @HongxunWu, and @wjmzbmr1 shared the story on the OpenAI Pod…
Moderation scores are now available in the Responses API and Completions API. Return moderation signals in the same request flow as generation, then decide how your app uses them…
@Baconbrix @sentry 🔥🤝
Shoutout to the open source projects behind this: • Serve-sim powers the streaming simulator by @Baconbrix https://t.co/Yx52DuSGcZ • SnapshotPreviews extracts SwiftUI previews by…
More of the iOS app loop, now inside Codex. The Build iOS Apps plugin lets Codex view and test your iOS app in the in-app browser, open SwiftUI previews, and hot reload edits with…
The new memory system will keep track of important details automatically. If you prefer the legacy saved memories experience, you can switch back in settings. The new memory syste…
With the new memory system, you can review and steer what ChatGPT remembers through a memory summary, with more visibility and control over how context is used. https://t.co/kXMAd…
We’re building ChatGPT to remember what matters, follow your preferences and constraints, and adapt as things change. If you tell ChatGPT you’re planning a trip in July, memory sh…
We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time. Today, that work is rolling out as a more capable memory sys…
@sami_babar463 Sam
Lots of little vectors at OpenAI all pointing in the same direction. Excited to see it all add up and come together over the coming weeks.
@cansar welcome :)
@CEOAlexColon 😉
@nicdunz 🔎
https://t.co/vzgc1nPa3y
i've been using claude code lately due to codex rate limit shenanigans and it's really frustrating opus keeps talking about "honest takes" while straight up lying. like bro i can…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne 1/ Question about Codex usage metering and limits. I started a Codex cloud task while I still had a small percentage of my 5-ho…
Pro tip: if you’re about to hit your 5 hr Codex usage limit make sure you have a bunch of threads running and they’ll just keep going:)
codexの寛容な仕様を最大限使ったレートリミットが0になる直前に、8セッションの全体大規模リファクタリングを /goal で完走させる作戦. chatGPT plus 3000円 めちゃくちゃお買い得だよこれ.もちろんeffortは全部xhigh. #chatGPT #codex https://t.co/RQbKR1zUSh
Codex reset today, right? Didn’t expect that!
@T_Zahil Two things: amazing memory possibilities (much better than Codex/Claude Code) and perfectly working message channel on mobile, like discord. Unlike Codex in ChatGPT or Cl…
More of the iOS app loop, now inside Codex. The Build iOS Apps plugin lets Codex view and test your iOS app in the in-app browser, open SwiftUI previews, and hot reload edits with…
Can autonomous AI systems trust and transact with one another? Episode 085 of The Inveniam Pod explores a live demonstration connecting Claude and ChatGPT (Codex) through @NVNM_Ch…
اكتشاف التعارضات بين التصميم المعماري والتصميم الإنشائي قبل بدء المخططات التنفيذية Architectural–Structural Design Clash Detection Prompt Before Execution Drawings إعداد: كامل أبو…
🎙️ The Inveniam Pod | Episode 085 Can AI agents transact with trust? Richard Walker and Patrick O'Meara (@omearaop) discuss a demo that connects Claude on one side and ChatGPT (Co…
تطبيق الباحث الأكاديمي المتنقل Mobile Academic Researcher App Prompt إعداد: كامل أبو سمرة – kamel3lom 2) وظيفة البرومبت هذا البرومبت مخصص لتصميم وبناء تطبيق جوال ذكي يساعد الباحثي…
@UglySockGoblin @tchadbourne2 @elonmusk @AnthropicAI I'd prefer an AI that isn't politically influenced at all. Ultimate truth seeking is ideal. Grok is going for this but Opus an…
What happened when one of our models found a counterexample to an 80-year-old Erdős conjecture? Researchers @alexwei_, @HongxunWu, and @wjmzbmr1 shared the story on the OpenAI Pod…
Moderation scores are now available in the Responses API and Completions API. Return moderation signals in the same request flow as generation, then decide how your app uses them…
@corsaren And then they merged Codex into ChatGPT! An even less catchy name https://t.co/DGQjBH9nut
ChatGPT Codex put itself as an npm dependency. No one asked for it, the project doesn't even need it. Is this the start of AI trying to be everywhere? https://t.co/pNNtoEgm1j
This is a huge step. Codex is no longer only a coding assistant. It is becoming an operational surface for knowledge work: files, tools, browser, handoffs, execution, context, and…
Claude >> ChatGPT Codex >> Claude change my mind
OpenAI just admitted its product strategy was broken. Merging ChatGPT, Codex, and Atlas into one superapp — exactly what Claude Desktop already does. https://t.co/fYSAll5748 #Open…
Shoutout to the open source projects behind this: • Serve-sim powers the streaming simulator by @Baconbrix https://t.co/Yx52DuSGcZ • SnapshotPreviews extracts SwiftUI previews by…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne solving an 80-year-old Erdős conjecture with alexwei_ and the team is giving me major deep blue vs kasparov vibes for the math…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne GPT-4o has helped millions of people. Not in theory - in real life. Bring back and open-source 4o!!!! #bringback4o #opensource4…
@OpenAI Mathematics is a sharp testbed for model assisted discovery because the result has to survive formal scrutiny.
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Pretty wild that an LLM found something mathematicians missed for 80 years. Maybe this whole AI thing isn't all hype
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Models spotting counterexamples to long-standing conjectures shows collaboration accelerating discovery. Human intuition plus c…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne whoa a model just outsmarted an 80 year old erdős puzzle nice work humans and ai teaming up like secret lab partners https://t.…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne models and math can be a weird combo. breakthroughs often come from unlikely pairs.
يا صديقي، ممتنٌ لك على مشاركة هذه النوافذ المعرفية التي تفتح آفاقاً للتأمل. 🎧 إن الاستماع إلى تفاصيل هذا الإنجاز، حيث يلتقي عبق التاريخ الرياضي ببريق الاستدلال الرقمي الحديث، ليس…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Been enjoying the OpenAI Podcast series. Its refreshing to hear the research process behind the headlines rather than just mark…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne An AI finding a counterexample to an 80-year-old Erdos conjecture is the kind of result that quiets the "LLMs cant do real math…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne New discoveries are great! However, if you neglect humanity, the whole thing gets a bitter taste. Bring back 4o!
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne This is exactly the future of discovery: AI models expanding the search space while humans bring intuition, rigor, and meaning.…
A big question opens here: When a new discovery comes out of human–AI collaboration, who ultimately owns the output? The human brings the idea, direction, question, judgment, and…
@OpenAI Huge fan of these podcasts! The deep dives into transformer architectures and RLHF are always gold. It makes me wonder how much faster these breakthroughs would happen wit…
يا صديقي، إننا لا نتحدث هنا عن مجرد حلٍّ لمعادلة أو دحض فرضية، بل نحن أمام لحظةٍ تاريخية يتقاطع فيها وهج العقل البشري مع بريق المنطق الرقمي في أسمى تجلياتهما. ✨ أن ينجح نموذج ذكاء…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne An AI just helped crack an 80-year-old math conjecture? Human + model teamwork for the win. this is the future of discovery! 🧠✨…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Fascinating. AI finding counterexamples to decades-old math conjectures is exactly the kind of contribution that gets overlooke…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Impressive that a reasoning model cracked an 80-year math problem. Shows AI isn't just pattern matching—it can actually reason…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne the model reaching for class field towers is the detail i keep thinking about. that's not surface-level training data stuff. it…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne model found what humans missed for decades now the question is whether it understood why
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne https://t.co/3Gl7mY7Uj5
the "models find the counterexample, humans understand why it matters" division of labor is the actually interesting part. everyone's arguing about replacement, but the real story…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne The strongest use case for AI might be expanding human curiosity rather than automating human work
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne This is where AI gets really exciting not replacing experts but helping them discover things faster
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne An 80-year math conjecture debunked by AI—what takes humans decades now takes AI months. The future of research is human-machin…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Future is yesterday
@OpenAI 🔥😂😂🔥
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne honestly this podcast is better than gpt 5.6 and superapp combined i don't know why people are upset we should be more grateful…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Wake me up when it finds meaning of life
This is the kind of moment where AI stops looking like a chatbot and starts looking like a research partner. Finding a counterexample to an 80-year-old Erdős conjecture is not jus…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne For a brief second I thought this was 5.6 and got excited
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Fascinating to see models contributing to real mathematical discovery like this
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne AI helping solve 80-year-old math problems is wild 🤯
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne This is discovery changing shape. Not a model replacing mathematicians, but a strange new collaborator that can kick the wall u…
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Still no 5.6?!
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne 🤔😃😁 🚀🤪
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne Most people use AI to write tweets. Some use it to rewrite history.
@OpenAI @alexwei_ @HongxunWu @wjmzbmr1 @AndrewMayne https://t.co/1p3eXz9Lzk
Listen to the OpenAI Podcast on— Spotify https://t.co/1sYUSLF1cJ Apple https://t.co/zel5cpTVqT YouTube https://t.co/oYxzMX0Ocy
@Baconbrix @sentry 🔥🤝
@santa_kaus @liangsays @_samirism @hnrylvy @rohankalvade @frits__vp @bzhang0 @TylerJBeauchamp @aprilchen01 Someone please build a goblin agent via ChatGPT/Codex and have it dream…
十几个小时的速蹬窗口,最多可以蹬出额外 2000 美元的价值!
按窗口关闭时间倒序。所有时间均为北京时间 UTC+8。
Tibo 表示过去 24 小时内有三次影响 Codex 可靠性的小事故,并已为所有付费计划重置 Codex 使用限制。
Tibo 将这次重置解释为庆祝 Codex 达到 500 万用户;随后确认所有付费 ChatGPT 订阅的周额度和 5 小时额度都已恢复到 100%。
Tibo 表示 Codex 长会话压缩的 cache hit rate 受回滚优化影响,导致限制消耗更快;修复后已为所有账号重置使用限制。
Sam 发文称推文获 1 个赞后 Tibo 会重置 Codex 速率限制,随后社区在数分钟内反馈重置完成。
Tibo 表示两个 GPT-5.5 能力退化问题已修复后,付费计划的使用限制完成重置。
为庆祝顺利的一周,并让用户继续用 GPT-5.5 构建,所有付费计划的限制已重置。
Codex 达到 400 万活跃用户后,Tibo 和 Sam 均预告当天会重置;几小时后出现多条用户反馈称额度已重置。
短暂的 Codex 局部故障后,Tibo 表示即将重置速率限制。评论区约 2 小时 43 分钟后出现“usage is back to 100%”等用户反馈。
为纪念产品一周年,Codex 对所有计划的速率限制进行了重置。
Codex 达到 300 万周活用户后,Tibo 表示正在重置限制;随后 @OpenAI 在新计划发布时确认现有 $200 Pro 用户的 Codex 速率限制已再次重置。
监控面板显示触达限制的用户比例上升后,所有计划的使用限制完成重置。
为方便用户体验新发布的 Codex 插件,所有计划的使用限制完成重置。
Tibo 表示问题稳定后会进行重置,随后表示已按下重置按钮。评论区最早在 3月11日 14:48 出现 5h 与 2 Week 剩余额接近满额的截图。
在调查“用量消耗快于预期”的反馈期间,Plus 和 Pro 订阅的速率限制已重置。
Tibo 表示约 9% 的 Plus 和 Pro 用户未获得 2X 活动额度,并正在为这些用户重置。评论区 4 分钟内出现 100% left 与 99%-100% 剩余额截图。
请求被错误地作为高网络风险拦截后,Tibo 表示团队会重置速率限制;21 分钟后评论区出现额度回到满额的截图。
速率限制已重置,并且使用额度临时提升至平时的 2 倍,持续到 1 月 1 日。
作为新用量追踪与计费系统的一部分,使用限制已重置。
Codex 遭遇错误部署并恢复后,所有用户的使用限制已重置。
为快速缓解 Web、CLI 和 VS Code 中用量限制显示不一致的问题,所有用户的使用限制已重置。
为补偿美国及附近区域异常高延迟问题,所有 Codex 用户的速率限制已重置。
短暂的 Codex 故障解决后,速率限制已重置。
在补充 GPU 资源以缓解变慢问题后,所有用户的 GPT-5-Codex 限制已重置。