I just canceled my ChatGPT subscription because I found an alternative that's 10 times better. Here’s how to use it: ↓

Wait... my phone just finished the deck I left half-done on my laptop. I closed my laptop mid-task. The work kept moving, and it remembered how I like it done. @kooko_ai #KooKoAI

Practicing presentations with classmates is helpful, but the feedback doesn’t always go beyond looks good. A realistic roleplay could push students to handle tougher questions, different audience reactions, and get much more actionable feedback.

Most AI demos feel like you’re interacting with a chatbot. Griffin feels different. The combination of vision, voice, and real-time reactions makes the interaction feel dynamic instead of scripted. It picks up on what’s happening and responds naturally as the conversation evolves. This is one of those AI experiences that’s much easier to understand once you try it.

AI is scaling faster than ever, and the infrastructure powering it needs to scale just as quickly. GMI Cloud raising $668M to expand GPU capacity across the U.S. and Asia-Pacific shows how much demand is building around AI training, inference, and production workloads. The race for AI infrastructure is only getting started. @GMI_Cloud @alex_yehya

Half a million downloads in a month. Today, our open source family takes another step forward. Thank you for the incredible support behind our first-generation models. We’re excited to introduce TwIL-LM3-Pro. At just 3.6 billion parameters, it brings powerful reasoning to everyday computers, with quantized builds that run locally. No cloud required. In our evaluation: Formal logic: Highest recorded headline score among the small models compared—beating China’s VibeThinker-3B by 35% and Qwen3.5-4B by 24%, and Liquid AI’s LFM2.5-8B-A1B by 47%. Broader reasoning: 95% on SVAMP and 64.1% on MuSR, the highest recorded scores among the small models compared. BIG-Bench Hard’s logic subset: 95.4%, compared with VibeThinker-3B’s 61.1%. We believe AI is entering a post-training era. The advantage will increasingly belong to companies with the best pipelines and those that can produce capable, personalized intelligence faster and more efficiently, then put it on devices people already own. That’s what we’re building at webAI. And we’re only beginning to share what’s coming out of our lab. Coming soon: Meridian, our family of frontier-class models built to run on device. Our most advanced models will be available through the @thewebAI application. Join the waitlist as we expand access. Proudly built in Austin, Texas. 🇺🇸

The part that stands out to me is how seamless the upgrade is. Ultravox’s built-in Inworld voices are now powered by Realtime TTS-2, while existing voice IDs continue working without any code changes. Developers get the improved voice experience without having to rebuild their integrations. That’s the kind of upgrade that actually matters in production.

Text, photo or video in. A fully rigged 3D character and its motion out. All in PINOC Agent Mode. No rigging, no mocap suit, no 3D skills. Export straight into your game. RT + comment "PINOC" and we'll DM you free credits 🎁

What I love about Eleven v4 Turbo is the voice keeping its energy through a long call.

The interesting bottleneck for coding agents may no longer be just writing code. It’s the feedback loop. Look at the visual → understand it → generate → render → inspect → identify what's wrong → fix it. PPTBench shows how difficult that loop still is, with only 2.57% of reconstructions passing every gate defect-free.

One thing I really like here: IFM isn’t just releasing the framework. They’re shipping the K2 Horizon checkpoints, training logs and recipes alongside it. That makes this much more useful for anyone who actually wants to understand how the models were trained.