AI just rolled out three major upgrades redefining speed, creativity, and omnichannel agents. From benchmark-crushing models to motion mastery and WhatsApp integration, heres whats new:

⚡ Gemini 3 Flash Drops — Googles Speedy AI Default Crushes Benchmarks
Google launches Gemini 3 Flash, a fast cost-effective model now default in the Gemini app and AI search worldwide, smashing scores like 33.7% on Humanitys Last Exam and 81.2% on MMMU-Pro while excelling in multimodal tasks, 3x faster workflows, and enterprise use via Vertex AI.

🎬 New Kling Motion Feature Out — Hyper-Realistic Video Control
Kling AI rolls out Motion Control to reference 30-second real videos and refine via text for lifelike movements, gestures, and expressions in demos like dances, battles, and fast action with zero jitter, building on Voice Control for pro filmmaking on Higgsfield.

🗨️ ElevenLabs Agents Add WhatsApp Support — Omnichannel Voice Chat
ElevenLabs Agents now integrate WhatsApp for single-agent deployment across web, mobile, phone, and messaging, delivering consistent voice/chat with full dashboard visibility, quick setup, and automated support to meet customers anywhere.

AI isn’t just evolving anymore, its powering speed, motion, and seamless connections.

Gemini 3 Flash Drops Googles Speedy AI Default Crushes Benchmark

Google launched Gemini 3 Flash, a fast, cost-effective model based on last month's Gemini 3, now the default in the Gemini app and AI search mode worldwide, replacing Gemini 2.5 Flash. It significantly outperforms its predecessor scoring 33.7% on Humanity’s Last Exam (vs. 11% for 2.5 Flash, nearing Gemini 3 Pro's 37.5% and GPT-5.2's 34.5%) and leading with 81.2% on MMMU-Pro excelling in multimodal tasks like video tips, sketch recognition, audio analysis, and visual responses with images/tables. Priced at $0.50/M input and $3.00/M output tokens (up slightly from 2.5 Flash but 3x faster, 30% fewer tokens for thinking), it's ideal for bulk workflows; enterprises like JetBrains and Figma use it via Vertex AI, with developer preview in APIs and Antigravity. Amid rivalry with OpenAI's GPT-5.2, Google processes 1T+ tokens daily on its API.

New Kling Motion Feature Out

Kling AI by Kuaishou rolled out Motion Control, enabling creators to reference real motion videos up to 30 seconds and refine them via text prompts for hyper-realistic body movements, sharp hand gestures, and perfectly synced facial expressions. Early demos impress with dramatic character performances, diverse poses, rooftop dances, anime battles, and high-speed action, all maintaining exceptional temporal stability without jitter. Building on recent Voice Control and native audio updates, this advances AI filmmaking tools for short-form content, ads, and dynamic animations, integrated seamlessly on platforms like Higgsfield for unlimited creative workflows. Creators praise its precision in fast action and complex choreography, making it ideal for professional-grade videos without traditional production crews, while unlocking new possibilities in AI-driven storytelling from viral social clips to cinematic sequences.​

ElevenLabs Agents Add WhatsApp Support

ElevenLabs expanded its omnichannel Agents platform to support WhatsApp, allowing teams to design a single AI agent once and deploy it seamlessly across web, mobile, phone lines, and the world's most popular messaging app for consistent voice and chat experiences. Customers can now speak or type questions via WhatsApp, receiving responses with the same reasoning, voice quality, and knowledge base used elsewhere, meeting users where they already communicate daily. The platform offers full visibility into all interactions through a unified dashboard for reviewing transcripts, analyzing performance, and updating behaviors that apply across channels, ensuring quality, compliance, and reduced operational overhead. Integration is quick via documentation, enabling immediate automation of support, issue resolution, and workflows on WhatsApp.

Hand Picked Video

BG Blur transformed from a basic blur tool into a powerful AI video enhancer with a minimal design, smarter background blur, privacy upgrades, and live previews for creators.​

Top AI Products from this week

• Shadow - Half of every meeting happens on screen, but most AI tools miss it completely. Shadow captures both: what was said and what was shown. Every word, every slide, every screen all without a bot joining your meeting. Shadow transforms that full context into real action. Write follow-up emails, extract action items, or create your own custom AI tasks. Never recap. Just move forward. 

• Brill - Brill is an iOS widget that helps you memorize a language on your own time. Cycling through the top 1000 words in a language, easily (re)view words in the language you're learning ,right on your home screen. Available in Spanish, French, German, and Dutch..

• Everything AI Tool - Everything AI Tool is built to dominate AI discovery. With 25,000+ curated AI tools and 30K+ unique weekly visitors, we give AI products real distribution, not launch-day spikes.

• TimeTuna.com - TimeTuna is on a mission to help 400 million people to enjoy scheduling and rescheduling.The first step: beautiful branded scheduling pages with gorgeous custom video backgrounds. If you care about aesthetics and design use TimeTuna. If you don't - use Cal.com or Calendly. Made in Amsterdam.

• ​Monocle for macOS - Monocle is a modern take on window dimming that elegantly blurs everything except your active window by simply shaking your cursor. It isn't just about productivity it's about presence. Feeling calm while you work, write, browse, or think. Designed to feel like it came with your Mac.

• LightBuddy- LightBuddy lives in your Mac’s Menu Bar and allows you to add an adjustable ring light around your display. Unlike the built-in feature Apple added in macOS 26.2, LightBuddy supports macOS 14 Sonoma or later, works on Intel and Apple Silicon Macs, and doesn't require an Apple-branded display.

This week in AI

• Arcads 2.0- Arcade raised $16M seed and launched Arcade 2.0 publicly after strong beta growth, enabling interactive product demos for marketers and PMs with video capture, team collaboration, and email exports.

• Tencent HY World 1.5- HY World 1.5 open-sources real-time interactive 3D world modeling at 24 FPS with geometric consistency, enabling game-like exploration via text/images and controls.

• Grok Voice Leads Speech Benchmarks- xAIs Grok Voice Agent tops Speech-to-Speech at 92.3% on Big Bench Audio with low latency and tool calling for $3/hour.

• Opal Enters Gemini - Google integrates Opal vibe-coding tool into Gemini web app for no-code creation of custom AI Gems via natural language and visual editor.

• ChatGPT Images Launches with GPT Image 1.5- OpenAI launched ChatGPT Images powered by GPT Image 1.5, its new flagship model, as a direct counter to Google's Nano Banana Pro, now rolling out in ChatGPT and via API for native image generation.

Paper of The Day

Researchers introduced Predictive Concept Decoders (PCDs), an end-to-end trained architecture that compresses neural network activations into sparse concept lists via an encoder, enabling a decoder to predict model behavior and answer natural language questions accurately. Unlike hand-designed interpretability agents limited by off-the-shelf models, PCDs scale effectively with data, improving auto-interp scores and downstream tasks like detecting jailbreaks, secret hints, implanted concepts, and latent user attributes. Pretrained on unstructured data and finetuned for specific queries, PCDs offer scalable, faithful explanations of complex activation spaces.

To read the whole paper 👉️ here