Jaxx

视频号@赛博吴同学

[Jaxx @jaxxwu]

全部都穿透出来了，真好用

[Nous Research @NousResearch]

Hermes Agent v0.23.0: The Hearth Release

Highlights:
- Desktop in a browser tab
- One gateway for every conversation
- /Rewind
- Group Chats with automatic routing
- Spend limits for paid tools
- Lighter and faster
and more

Changelog below

[Jaxx @jaxxwu]

@QwenDevs 你们营销就是吊胃口么，都 24 点了，要睡觉了

[Jaxx @jaxxwu]

@MiaAI_lab 都下班了

[Jaxx @jaxxwu]

@theo 一周又一周～

[Jaxx @jaxxwu]

刚刚刷到 DeepSeek 那篇，本来手指头都准备划走了，结果头两行数字给我整精神了：日活 154 万，一天啃掉差不多 92 万亿 token，第三方插件攒到 3.5 万个，活跃用户里约莫六成自己装过、甚至自己糊过一个。好家伙，这哪是发了个桌面端，这是把摊子往街边一支，吆喝大家自己上来摆货啊。
文中提到的几位作者。蜗牛把记灵感、写脚本、做封面、画分镜一股脑塞进一个工作台，原话是"睡醒、打开 DSH、秒开工作台、开蹬"——这不就是咱每天早上开机那股莽劲儿么。歸藏把设计那套 Skill 搓成插件，一篇文章能变网页、变卡片、变封面，还能点"做同款"，说作品、专栏、语气都能攒下来，出活儿就稳了。另外俩，一个专门让你看清 Agent 到底嚼了哪些上下文、怎么随对话变，一个把读论文、划高亮、翻译、记笔记串成一条线，全是自己用着别扭、干脆撸袖子自己整的。
说白了官方不想啥都自己兜着，专业活儿谁最懂谁上。这路子我挺吃，比闷头憋个大版本有意思。就是好奇那三万五千个插件里，有多少是装完当天就再没点开过的😂

[Jaxx @jaxxwu]

mark

[Eian @EianLu]

https://t.co/IjmUF2hOCn

[Jaxx @jaxxwu]

@_catwu 哈哈

[Jaxx @jaxxwu]

Hy4 这次拿了 35 分，跟 Grok 4.7 打平。模型榜看多了有点像看高考排名，差一分能吵一整天；可真用起来，价格、速度、会不会瞎编又是另一张卷子。分数我先记下，等有空拿它干点真活再说。毕竟榜单上夺冠是一回事，别让我改三遍提示词是另一回事。

[Artificial Analysis @ArtificialAnlys]

Hy4 from Tencent scores 35 on the Artificial Analysis Intelligence Index v5 and is the #2 model from a China-based AI lab

Hy4 is the latest model from @TencentHunyuan and is the successor to Hy3 released earlier this year. It scores 35 on the Artificial Analysis Intelligence Index v5, equal with Grok 4.7 (xhigh) and 1 point behind Claude Haiku 5.5 (max). Hy4 is the #2 model from a China-based AI lab, behind only Kimi K3.1 Pro, which scores 46. Hy4 is priced at $0.834/$2.501 per 1M input/output tokens.

Key takeaways:

➤ Hy4 excels at agentic coding tasks. Hy4 scores 31% on FrontierCode 2.0, compared to 32% for Claude Fable 5.1 (max). It is the leading model from a China-based AI lab, with DeepSeek V4.1 Flash 1017 (max) scoring 28% and Kimi K3.1 Pro scoring 25%. On Terminal-Bench 4.0, it scores 37%, the fourth-highest score among models from China-based AI labs

➤ Hy4 outperforms its peers on knowledge and reasoning. Hy4 achieves accuracy of 50% on AA-Omniscience v1.1, 1 point ahead of Kimi K3.1 Pro, but has a high hallucination rate of 89% compared to Kimi K3.1 Pro’s 37%. On HLE-Diamond, it scores 33% compared to Kimi K3.1 Pro’s 37%

➤ Hy4 offers a competitive Capability vs. Cost trade-off compared to to peers. With a Cost per Intelligence Index Task of $0.78, Hy4 (35) is 10-times cheaper than Grok 4.7 (xhigh, 35) and 13-times cheaper than Qwen3.8 Max (0902) (34) but almost twice as expensive as Claude Haiku 5.5 (max, 36)

Other model details:

➤ Context window: 1 million tokens

➤ Pricing: $0.834/$2.501 per 1M input/output tokens. Cache reads $0.042 per 1M tokens

➤ Multimodality: Text and image input, with text output

[Jaxx @jaxxwu]

这次更新值得关注的，不只是“能做 GIF”，而是把静态图像生成推进到可复用、可延展的视觉工作流：提示词、图片和角色参考能连续迭代，创作门槛会继续下降。接下来真正拉开差距的，可能不是单张图有多惊艳，而是生成过程是否稳定、细节是否可控、素材授权与肖像边界是否清楚。工具越方便，越需要把创意和责任一起设计进去。

[ChatGPT @ChatGPT]

We have a gif for you—ChatGPT Images 2.6 is here.

ChatGPT Images 2.6 can turn your prompts and pictures into animated GIFs.

You can also keep reference images of yourself, pets, and characters ready to reuse in new creations...no need to upload them again.

Go on. Make it weird. Send more GIFs of your dog dancing.

Images 2.6 is available today in ChatGPT, and Codex across all tiers on web, desktop, and mobile.

https://t.co/WQg0cVJRca

[Jaxx @jaxxwu]

@_FORAB 内部 AI 额度也砍啊 挺离谱

[Jaxx @jaxxwu]

Meta 收购 Conductor 了，并行跑多个 AI coding agent 的那家。CEO 带着团队进 Superintelligence Labs。现在卷的不是再出一个模型，是怎么同时放一群 agent 干活还不乱。编排层开始值钱了

[Techmeme @Techmeme]

Meta acquires Conductor, which lets developers run multiple AI coding agents in parallel; CEO Charlie Holtz says the team will join Meta Superintelligence Labs (@madisonmills22 / Axios)

(Visit Techmeme dot com for the link and full context!)

[Jaxx @jaxxwu]

@Stargridcompass 烧钱烧到最后还得合并啊 活久见

[Jaxx @jaxxwu]

@laopai1990 77倍啊卧槽 排行榜他们根本不看

[Jaxx @jaxxwu]

Claude Code 后台 subagent 刷状态，从整段对话重读改成 Haiku 拿短笔记写，单次大概省 99%。

agent 光汇报「我还活着」就能把额度啃掉一块，挺离谱。这笔账总算抠回来点了

[ClaudeDevs @ClaudeDevs]

Background subagents in Claude Code now use less of your usage limit.

A running subagent updates its status line every 30 seconds, and each update used to re-read its whole conversation. Now Haiku 5.5 writes it from short notes instead, so each update costs about 99% less.

[Jaxx @jaxxwu]

@laopai1990 学会划水了啊 挺离谱

[Jaxx @jaxxwu]

Cat Wu 跳 OpenAI 了啊

Claude Code 做起来的那个人，现在去带对方 Core Products

coding agent 这仗打到这份上挺无语的，模型差几分大家不在乎了，挖产品负责人比发新模型还炸

[Jaxx @jaxxwu]

@Stargridcompass 眨眼这效果绝了

[Jaxx @jaxxwu]

@aizhibeibei agent 框架直接上云了啊 沙箱还挺便宜

[Jaxx @jaxxwu]

@gdb outbound telephony from live agents damn

[Jaxx @jaxxwu]

tok/s 这指标要翻车了。流得飞快但写一堆废话，任务反而更慢。Artificial Analysis 改成 Time per Task，Astra 51 tok/s 完任务比 Opus 111 还快。agent 时代拼的是少说话多干活吧

[Artificial Analysis @ArtificialAnlys]

Announcing the next version of Artificial Analysis API Performance Benchmarking: Time per Task is now our headline performance metric, measuring how long a model takes to complete a task rather than how fast it streams

The update applies to the ~1,000 serverless endpoints we benchmark multiple times a day for speed, latency and caching.

Key changes:
➤ Time per Task is now the headline metric to compare models: as workloads become more agentic, the time to complete a task matters more than speed per token. A model that is fast per token but verbose can take longer than a slower model that uses fewer tokens. Time per Task divides a model's output tokens per Intelligence Index task by its End-to-End Output Speed.

➤ End-to-End Output Speed covers the full request: speed is now measured from sending the request to receiving the final token. That includes the wait before the first token (prefill, and reasoning for models that hide it) as well as generation. Output tokens now come from each provider's API-reported counts instead of the o200k_base tokenizer, and we monitor for discrepancies in reported counts.

➤ Agentic workloads are now part of every measurement: we benchmark short, document-length and long-context prompts, plus multi-turn sessions across coding and knowledge work. Headline figures are the average across these workloads over the past 5 days.

➤ API default effort is now tracked alongside highest effort: most developers run models at the effort level each lab sets as its API default, so we now benchmark key models at that setting as well as at their highest effort. Effort names are not comparable across labs, so we use each lab's own default rather than one shared setting.

Key takeaways:
➤ OpenAI models make up most of the Intelligence vs. Time per Task Pareto frontier despite slower output speeds: across effort levels, GPT-6 Astra and GPT-6.1 Sol hold the frontier from ~45 to ~55 on the Intelligence Index. At max effort, GPT-6 Astra runs at 51 tokens/s but finishes a task in ~17 minutes, against ~26 for Claude Opus 5.5 (max, 111 tokens/s) and ~29 minutes for Gemini 4 Carbon (high, 71 tokens/s), because it uses ~50k output tokens per task against ~120k to ~170k.

➤ API default effort cuts Time per Task by ~3 to 5x and Cost per Task by ~2 to 4x for GPT-6 Astra, GPT-6.1 Sol and Claude Opus 5.5: all three default to medium effort, which scores ~5 to 11 points lower on the Intelligence Index in exchange. Claude Opus 5.5 falls from ~26 minutes and $10.04 per task at max to ~6.8 minutes and $2.45 at medium effort.

➤ Endpoints that optimize for decode speed alone see the largest drop in End-to-End Output Speed: high decode speeds can come at the cost of latency, for example by queuing requests to run at lower batch sizes, or by holding output and sending it in a single burst. Both lengthen the wait before the first token, which End-to-End Output Speed now counts.