I’m always interested in seeing how far AI can go beyond generating code. IQuest-Q1 is a code-focused large model built for developers, with performance evaluated across different task scenarios. From building interactive game demos and generating 3D scenes to debugging code and handling complex tasks with multimodal information and tools, it explores some interesting possibilities for AI in development. Excited to see how models like this evolve! Explore IQuest-Q1: iquestlab.github.io/ #coding #LLM

Remember the mysterious Space Bunny model everyone was curious about? We finally know what's behind it: MiniMax M3.1 Flash! In this video, I compare its visual outputs with DeepSeek V4.1 Flash 1017, using examples of animated scenes and creative designs. Both deliver similar results in these examples, making it interesting to see how they approach the same prompts. What I find most interesting is the balance between output quality and cost. Beyond creative tasks, M3.1 Flash is built for coding and agentic workflows, offering 70% lower overall cost for agentic tasks compared to its predecessor, with output speeds reaching 200+ tokens per second.

The next big challenge in AI isn’t just building smarter models. It’s making them respond faster. As agents reason, use tools, and make repeated model calls, latency adds up quickly. Memory bandwidth becomes a key factor in generating tokens efficiently. That’s why @general_compute partnering with @positron_ai is worth watching. Atlas inference systems are expected to be available in Q1 2027, with next-gen Titan systems to follow. Matching the right hardware to each stage of inference could be key to scaling the next generation of AI agents.

Our heterogeneous cloud just got better. We're announcing a cluster scale deployment of @positron_ai as our next partner. We believe demand for fast inference will outstrip supply for the next decade. Our bet on Positron is another step towards getting fast inference for opensource models into builders' hands. Hardware goes live Q1 2027. Thanks to @mitesh711, @trsohmers and the whole Positron team! We're excited for this partnership. LFG!

We're living in the worst era of design. AI made creating visuals easier, but somehow, everything looks the same. Vyb takes a different approach. It builds designs layer by layer using your actual brand assets, keeping your logos, fonts, colors, and product photos intact. You can edit individual elements and export your designs to Figma. Finally, you can create marketing visuals that feel like your brand, without spending time fixing generic AI designs.

I launched the first image and video generation models at Google, and kept asking why all AI designs look the same. AI has brought us the WORST era of design. It made content creation effortless, but it also made 90% of it look painfully generic. My co-founder and I spent 2 years building in stealth to change that. Meet Vyb, the first AI design model that understands your taste like a designer would. RT + comment your website URL and we’ll make your first design.

I'm especially interested in the pages where handwriting, formulas and tables all show up together. That's where I'd point OXR first. It covers those cases with structured Markdown/JSON and source coordinates, and at 1.3B parameters it still comes out on top of olmOCR-Bench, OmniDocBench and DocParsingBench. On speed it's the quickest of the open-source options tested, with 2.52× MinerU2.5-Pro's throughput on one RTX PRO 6000. Keeping layout analysis replaceable should also help when different projects need different element groupings. github.com/SoMarkAI/OXR

Beyond OCR. Meet OXR. Only 1.3B. Open source. #1 on olmOCR-Bench, OmniDocBench & DocParsingBench🏆 #1 in parsing speed among open-source models⚡ RSI-powered. A first in document parsing. Turn documents into Agent-Friendly data. If OXR helps, give us a star on GitHub! 🌟

Even revolutionary technology can lead to a massive market bubble. Ben Horowitz makes a strong case for why this AI boom is different from the dot-com bubble. But that doesn’t mean every bet on AI will pay off. The real question is whether the market is pricing in more than AI can deliver.

On this weeks @theallinpod @chamath @davidsacks @jason spoke with Ben Horowitz from @a16z about how this buildout is different from the internet bubble buildout in the late 90s. Ben’s take is smart and he identifies a real difference. That doesn’t mean he’s correct. I spent some time thinking through HOW Ben might be wrong about this. NOT a prediction, but well worth discussing the risks.

The future of home robotics is looking more exciting than ever! I love how Tuesday Labs is thinking beyond what a robot can do and focusing on what it’s actually like to live with one. Voxel combines practical capabilities, thoughtful design, and the charm of a little gentleman in training. Having a robot that can tidy up, find misplaced items, and help with everyday tasks could make life at home so much easier!

Introducing Voxel, a droid butler made by Tuesday Labs.

1/ Ever had an AI agent suggest an approach your team had already rejected? The problem isn't always the model's ability to write code. Sometimes, it lacks the context behind your team's previous decisions. I wanted to explore how EverOS Cloud could help solve this by giving agents access to past project experience.

Introducing Zavi. Growth on autopilot. ✨ Give it your website. Zavi finds what’s holding your growth back, then runs your ads, SEO, social posts, and emails.

Most AI agents are just chatbots with a dashboard. This one runs ads, fixes SEO, optimizes funnels, publishes posts, and emails customers. Useful or a little scary?