Imagine Video 1.5 launched last month as our best video model yet, with more lifelike motion and sound. Today it goes even further with text-to-video support, image and voice references, and native 1080p.
grok.com/imagine
连环推 ×7This month’s Gemini Drops brings even more intelligence and agentic power to your desktop and mobile experience.
Let's dive into what's new in this 🧵
1) Use your voice in the Gemini app for macOS
Dictate, summarize files, and rewrite copy directly into any window with your voice in the Gemini app for macOS.
2) Access Gemini Spark in more countries and languages
Now even more people across the world can use Gemini Spark as their 24/7 personal AI agent to get things done more efficiently.
3) Try Gemini 3.6 Flash & 3.5 Flash-Lite
Upgraded versions of our Flash and Flash-Lite models are now available, bringing better reasoning at greater speeds.
Try them by selecting “3.6 Flash” or “3.5 Flash-Lite” from the model dropdown menu at
gemini.google
or in the app.
4) Set up your digital avatars to use with Nano Banana
Set up your digital avatar in Gemini once, and you can seamlessly create custom images or videos of yourself without having to upload a selfie as your reference image.
5) Connect more apps in Gemini to make your life easier
Now you can connect even more apps in Gemini like
@Dropbox
,
@Viator
, and
@Zillow
to easily organize your files, book tours and activities, and search for your next apartment rental.
6) Create personalized images in Gemini
Use Gemini to create deeply personalized images based on your unique interests and preferences. Now available to all users in the U.S.
Gemini Spark is rolling out to Google AI Pro users outside the U.S.
Spark is your personal AI agent that works in the background 24/7 to get things done under your direction, handling the heavy lifting so you can focus on what matters.
引用 Google Gemini @GeminiAppGemini Spark is now also starting to roll out to Google AI Pro subscribers in the U.S., and we'll be bringing it to AI Pro subscribers in more countries soon. More info on where Gemini Spark is available here: goo.gle/4fc7349查看被引原帖 ↗
Thinking Machines@thinkymachines · 公司官方 · 9 天前
Releasing weights indiscriminately isn't safe. Neither is keeping capable models inside a few labs.
We think there's a path between them. We haven't mapped all of it. Our new post covers the part we can see: how we assessed Inkling, and why access should widen in stages.
thinkingmachines.ai/blog/a-s…
what a privilege it is to be tested
DeepSeek-V4-Flash-0731 is now available on Ollama's cloud. This update substantially enhances the model's agentic capabilities:
ollama run deepseek-v4-flash:0731-cloud
Use it with Claude Code:
ollama launch claude --model deepseek-v4-flash:0731-cloud
A conversation with the Google DeepMind Robotics team about our latest Gemini Robotics 2 model, the arc of robotics progress, and where we go next 🤖
More info on where Gemini Spark is available here:
goo.gle/4fc7349
idoubi@idoubicc · 中文博主 · 9 天前
开源 DSCode：为 DeepSeek 深度优化的 Coding Agent
核心卖点：快、省、稳、开放、安全
- 快：DeepSeek V4 Flash + 极简 Agent Harness，原生适配 Responses API、推理档位、流式响应、代码 Patch 和服务端 Web Search，减少工具选择与协议开销。
- 省：针对 1M Context 和前缀缓存优化，提高长会话缓存命中率；可实时查看 Token、缓存命中和费用，让每一笔模型开销都清清楚楚。
- 稳：内置 Plan、会话恢复、Checkpoint/Undo 和后台任务；支持多 Agent 并行，并通过独立 Git Worktree 隔离修改。
- 开放：支持 Skills、Hooks、MCP；DeepSeek 优先，也可以切换 GPT、Claude 等模型。
- 安全：API Key、配置和会话默认保存在本地；命令运行在 OS Sandbox 中，工具权限由用户控制，Provider 密钥不会传给子进程。
DSCode Agent Runtime 基于 Pi Agent Toolkit，采用 MIT 协议完整开源，可以自由 Fork、修改，也可以集成到你自己的 CLI、IDE 或桌面产品中。
欢迎试用，感谢 Star ⭐
开源仓库地址👇
And you can watch on YouTube:
youtube.com/watch?v=-rYFDefc…
: )
Thinking Machines@thinkymachines · 公司官方 · 9 天前
We’re also expanding our safety team. If you’re interested in joining, please reach out!
thinkingmachines.ai/#join-us
That’s a wrap on this month’s Gemini Drops!
To learn more about everything that’s new, check out the Gemini Drops Hub:
gemini.google/gemini-drops
Matt Wolfe@mreflow · 博主 · 9 天前
I have this weird nostalgia for when AI was more difficult to use. Back when we had to use GPT-3 purely through the API or GitHub Copilot and you practically needed to know how to code to run an image generator…
The struggle and the wonkiness was what made it fun. Easy is so much less fun.
九原客@9hills · 中文博主 · 9 天前
即使不包含任何自我认知数据，用Claude合成的数据微调后，就容易自认为Claude。
尤其是 SFT 方法，会大幅调整分布，这个实验有点意思。
引用 Ziqian Zhong @fjzzq2002New blog: Why does Kimi identify as Claude and Sonnet 4.6 as DeepSeek? We find that this can come from a subliminal-learning-like effect: If you speak like Claude, you become Claude.查看被引原帖 ↗
Bilawal Sidhu@bilawalsidhu · 博主 · 9 天前
One day later and nano banana inside google earth has been rolled back.
Some say earth is the evidentiary layer and no place for gen ai slop.
Others say it’s a useful filter on reality and misinformation is already a screenshot away.
What’s ironic is this imagery was watermarked, private and never graduated to a factual layer.
Rolling the feature back pending further generation restrictions satisfies neither camp.
引用 News from Google @NewsFromGoogleRe: our statement on Image Generation in Google Earth: “We know that people uniquely trust Google Earth for a reliable view of the world. We’ve seen geospatial professionals using this feature for a range of useful purposes, however we’ve also seen people sharing screenshots of generated imagery that appear to violate our policies. So we’re rolling back this feature in Google Earth while we work on implementing stronger guardrails. It's important to note that generated images didn’t appear in the main Google Earth experience for others to see and were watermarked as AI generated.”查看被引原帖 ↗
Bilawal Sidhu@bilawalsidhu · 博主 · 9 天前
The Call of Duty heartbeat sensor is real.
Radio waves were built to carry information. Now AI is learning to read the physical world hidden inside their reflections.
I trace the lineage of RF sensing from to the Army-funded Sense Through the Wall program and Biometrics at a Distance -- plus research that pulls breathing, heartbeat, pose and identity out of thin air.
The same physics now drives $30 radar chips and telco patents for scanning environments with the cellular network itself: 5G/6G ISAC aka. integrated sensing and communication.
The goal is mobile infrastructure that senses drones, robots, traffic and weather -- a real-time map of the physical world. No cameras required.
Chapters:
00:00 The heartbeat sensor is real
00:53 How radio sees you
01:12 Sense Through the Wall
02:12 Biometrics at a distance
02:46 Your heartbeat is an ID
03:28 Five layers of RF sensing
07:10 Radar on your bedside table
08:42 Why India blocked the Pixel 4
09:14 From phone radar to city-scale sensing
10:17 The $22 version
11:33 The Ghost Murmur mystery
13:48 What radio waves really see
Try it today:
app.runwayml.com/
idoubi@idoubicc · 中文博主 · 9 天前
官网地址：
dscode.ai/
一行命令快速开始：
npm i -g @thinkany/dscode
idoubi@idoubicc · 中文博主 · 9 天前
github.com/thinkany-ai/dscod…
StepFun@StepFun_ai · 公司官方 · 9 天前
Voice AI Night was 🔥
Packed room in Mountain View last night for our keynote on next-gen end-to-end speech models + a founder panel with
@PLAUDAI
@cresta
and Firework on building voice AI at global scale.
Thanks Seamate for co-hosting — and thanks to everyone who came out, ate, and asked great questions 🎙️
More builders, more voice AI, soon 👀
Bilawal Sidhu@bilawalsidhu · 博主 · 9 天前
Here's the youtube version w/ chapters + sources if you prefer that. Enjoy!
youtube.com/watch?v=olaQ3-m2…