AI 热点

重置监控

盯着 Google 的 Varun 公布的 Antigravity 额度重置:什么时候生效、谁能用上、原话怎么说。

全部为北京时间 · UTC+8

上一次额度重置:11月21日确认

当前没有等待生效的重置

确认帖 11/21 17:18 · 不是精确到账时间

不预测尚未宣布的下一次重置。Varun 一旦宣布,这里会显示预计生效时间与原帖。个人额度请在 Antigravity 内查看。

Varun@_mohansolo确认完成

已重置所有人在 Antigravity 上的 Gemini 配额。 缓存改进使你在配额内可完成的工作量增加了 22%。周末愉快地使用 Neon 构建吧!

英文原文

Reset everyone’s Gemini quota on Antigravity. The caching improvements have increased the amount you can get done in quota by 22%. Enjoy building with Neon over the weekend!

近 90 天额度重置
7 次
重置间隔中位数
9.7 天
上次额度重置
11月21日

重置日历

点任意日期查看当天的重置、发卡和原帖。

浏览 15 个历史记录日期
2091年10月

本月 0 次额度重置 · 0 次等待生效 · 0 次待确认

已确认应已生效,未确认已过预计时间,待确认已宣布,等待生效
最近相关原帖33 条公开记录
Varun@_mohansolo

@huang18825 将在下一个版本中修复,这个问题被我们疏漏了。

回复与引用上下文(1 条)
回复 @huang18825

作为 Antigravity 的重度高阶用户,我对最近的应用更新以及众多强大的 Gemini 新模型陆续上线感到由衷兴奋——团队干得太棒了! 不过,我今天深入研究了一个令人困扰的问题:language_server.exe 在启动时通过 Init -> PruneAll(500) 硬编码了 500 条聊天记录的上限,会悄无声息地清除较旧的本地历史记录,而且 config.json 或 settings.json 中完全没有配置选项。由于每个子代理都会生成自己独立的对话,任何使用 Teamwork 或 Swarm 模式的人几乎都会立即触及这个 500 条上限!更糟的是,被清除的聊天记录仍然显示在界面侧边栏中,但点击时无法加载(“Conversation unavailable”)。 能否让这个上限可配置,并同步界面状态? @antigravity @rodydavis

英文原文

@huang18825 Fixing in next release, this slipped through the cracks.

Varun@_mohansolo

@shneural 是的,对于模型调用之间间隔较长的工作负载,收益会大得多。

回复与引用上下文(2 条)
回复 @shneural

Google 真的修复了 Antigravity 中 Neon 的缓存!他们说使用量会增加约 22%,但我现在的额度实际上大了 3–4 倍。 非常感谢 @_mohansolo 和团队响应如此迅速。我太喜欢 Neon 了,甚至订了两个 AI Ultra 20x 套餐来全职使用它。DeepMind,你们太棒了!! https://t.co/UDjPVRghUp

引用 @shneural

Gemini 4 Neon 消耗 Antigravity 额度的速度快了约 3 倍。提示词缓存坏了(1 分钟后命中率为 9%)。 Opus 5.5:约 95%(1 小时) Codex 中的 GPT:约 97%(24 小时) 很糟,因为 Neon 用于智能体真的强得离谱。Google,我知道现在基础设施压力很大,但请修复一下(让 Neon 来修吧,哈哈,再重置一下额度) https://t.co/sBqP9RprUt

英文原文

@shneural Yeah, for workloads where time between model Invocations are high, it will be a much bigger win

Varun@_mohansolo确认完成

已重置所有人在 Antigravity 上的 Gemini 配额。 缓存改进使你在配额内可完成的工作量增加了 22%。周末愉快地使用 Neon 构建吧!

英文原文

Reset everyone’s Gemini quota on Antigravity. The caching improvements have increased the amount you can get done in quota by 22%. Enjoy building with Neon over the weekend!

Varun@_mohansolo

@OrganicGPT 我们正在解决这个问题!

回复与引用上下文(1 条)
回复 @OrganicGPT

Cache hit rate >80% in 2026 is pretty standard. I Get +99% on DeepSeek Harness and V4.1 Flash. Unfortunately, Gemini 4 series has abysmal hit rate https://t.co/jfP2swCw8i

英文原文

@OrganicGPT We’re working on it!

Varun@_mohansolo

@aicodeking 应该很快就会修复!https://t.co/3vRHpOXVvr

回复与引用上下文(3 条)
回复 @aicodeking

I investigated this a bit and it seems that the cache TTL for Antigravity is somewhere near 5mins. It keeps hitting 90-98% cache until ~5 mins of idle time and then it goes downhill to 9%. I haven't noticed this because of long-horizon tasks but it does seem to happen. The 15-30s or 30-60s or 1-2mins claim is not correct for me. It only ever happens after 5mins. But, this shouldn't happen as TTL on the Gemini API sits at about 1hr and Claude, Codex holds about 85% throughout that period as well.

引用 @_mohansolo

We’re rolling out a change to increase cache hit rate on Neon by around 4% on Antigravity by massive increasing the cache pool size. This will increase what you can get done in quota by at least 20%. Will issue a reset after it’s done since it burned quota faster!

引用 @theo

Once the honeymoon phase wears off a lil, I’ll go back to harassing Google into fixing these things. Neon is the first Gemini model that’s good enough to actually deserve a functioning platform.

英文原文

@aicodeking Should be fixed soon! https://t.co/3vRHpOXVvr

Varun@_mohansolo

更多信息:对于模型调用之间间隔时间较长的工作负载,命中率尤其糟糕(缓存已被逐出)。这项改动应该能解决该问题,并让这些工作负载看到幅度大得多的改善。 我们会继续在全球范围内改善这一指标,给大家带来不便,敬请谅解!

回复与引用上下文(1 条)
回复 @_mohansolo

We’re rolling out a change to increase cache hit rate on Neon by around 4% on Antigravity by massive increasing the cache pool size. This will increase what you can get done in quota by at least 20%. Will issue a reset after it’s done since it burned quota faster!

英文原文

More info: The hit rate was particularly bad for workloads where there was more time between model invocatikns (cache was evicted). This should fix that and they should see a much higher improvement. We will continue to improve this metric globally, sorry for the inconvenience!

Varun@_mohansolo预告

我们正在 Antigravity 上推出一项改动,通过大幅增加缓存池容量,将 Neon 的缓存命中率提高约 4%。这将使你在配额范围内至少多完成 20% 的工作。 完成后会发放一次重置,因为它消耗配额的速度更快!

英文原文

We’re rolling out a change to increase cache hit rate on Neon by around 4% on Antigravity by massive increasing the cache pool size. This will increase what you can get done in quota by at least 20%. Will issue a reset after it’s done since it burned quota faster!

Varun@_mohansolo

@shneural 私信聊?这情况相当出乎意料,很想深入调查一下。

回复与引用上下文(1 条)
回复 @shneural

Gemini 4 Neon is burning your Antigravity limits ~3x faster. prompt caching is broken (9% hit after 1m) Opus 5.5: ~95% (1h) GPT in Codex: ~97% (24h) sucks cuz Neon is actually insane for agents. Google i know infra is rough rn, but pls fix (put Neon on it lol + reset limits) https://t.co/sBqP9RprUt

英文原文

@shneural DM? Would love to dig since this is quite unexpected

Varun@_mohansolo

@aicodeking 有意思,它不应该如此频繁地轮询。模型应该设置一个相当合理时长的计时器并使用它。

回复与引用上下文(1 条)
回复 @aicodeking

Gemini Neon is really good.. Just one feedback... This is probably a harness related one.. A long command can only stay in the foreground for 10 seconds. After that it goes to the background. To wait for it, manage_task status needs WaitDurationSeconds. 0 means "tell me now." 300 means "wait until it finishes." That field is required. If you leave it out, agy does not say the call is wrong. It answers "still running" right away, same as 0. The agent then asks again on every turn. I hit this on a 30 minute render. 195 status checks in 15 minutes. The render was still going at about 48% when the session died. There were 657 status checks in that session. Please reject a status check that has no wait time, or treat a missing wait as 300, not 0. Neon does this more often when using via ACP. @_mohansolo @OfficialLoganK please fix this if possible.

英文原文

@aicodeking Interesting, it should not be polling this much. The model should set a fairly reasonable length timer and use that.

Varun@_mohansolo

@wolfie_ 奇怪,你能分享一下错误 ID 吗?

回复与引用上下文(2 条)
回复 @wolfie_

@_mohansolo Neon High

回复 @_mohansolo

@wolfie_ Hmm, what model are you using?

英文原文

@wolfie_ Strange, can you share your error ID?

Varun@_mohansolo

@wolfie_ 嗯,你用的是什么模型?

回复与引用上下文(1 条)
回复 @wolfie_

antigravity is unusable, can't go more than a few minutes without hitting "Agent execution terminated due to error"

英文原文

@wolfie_ Hmm, what model are you using?

Varun@_mohansolo

@BenjaminDEKR 你遇到的超时具体是什么情况?@shengzheyao

回复与引用上下文(1 条)
回复 @BenjaminDEKR

Google: nice models, guys. Now please fix your CLI harness so it stops timing out and is actually usable! 🙏

英文原文

@BenjaminDEKR What are you seeing in terms of time outs @shengzheyao

时间是怎么推算的?来源与规则

有原话就按原话。原帖写了钟点、next hour、end of day 等时间时,按美国太平洋时区换算为北京时间,并保留有限的传播延迟。

只写日期时按当天太平洋时间全天估计;没有写时间时暂按发帖后半天估计,不假定存在固定重置周期。

预计结束后六小时内显示待确认,之后显示应已生效,仍不代表官方确认。确认帖日期不代表精确到账时间。

监控计划每 10 分钟检查公开帖子。是否已完成检查以监控状态为准;适用范围以原帖为准,个人额度请在产品内查看。

历史记录始于 2026-05-21
需要关注 · 最近检查 11/22 02:55AI News 整理 · 非 Google 官方页面
上次尝试读取
11/22 02:55
上次读到帖子
11/22 02:55
上次完整检查
11/22 02:55

最近相关帖子提到了产品故障。查看原帖