Hugging Face unveils two new humanoid robots

✨TodayOnAI’s Daily Drop

• Hugging Face unveils two new humanoid robots

• DeepSeek’s distilled new R1 AI model can run on a single GPU

• Perplexity Launches Labs to Take AI Beyond Search

• 💬 Let’s Fix This Prompt

• 🧰 Today’s AI Toolbox Pick

Hugging Face

Hugging Face unveils two new humanoid robots

🚀 TodayOnAI Insight: Hugging Face is deepening its robotics ambitions with two new open-source humanoid robots—HopeJR and Reachy Mini—designed to democratize access to robotic development and experimentation.

🔍 Key Takeaways:

• Two new robots launched: HopeJR is a full-size humanoid with 66 degrees of freedom; Reachy Mini is a compact desktop robot with interactive capabilities like speech and motion.

• Open source and affordable: HopeJR is expected to cost ~$3,000; Reachy Mini ~$250–$300, depending on tariffs.

• Available later this year: A waitlist is open now, with initial units expected to ship by year-end.

• Powered by acquisition: The robots are a result of Hugging Face’s April acquisition of Pollen Robotics.

• Ongoing robotics expansion: Builds on Hugging Face’s LeRobot platform and recent updates to its SO-101 robotic arm.

💡 Why This Stands Out: Hugging Face is challenging the high-cost, closed-source status quo of robotics with tools that are not only programmable but accessible. Its strategy underscores a broader shift toward community-driven innovation in embodied AI—where open ecosystems, not just large labs, shape the future of intelligent machines.

DeepSeek

DeepSeek’s distilled new R1 AI model can run on a single GPU

🚀 TodayOnAI Insight: DeepSeek quietly released DeepSeek-R1-0528-Qwen3-8B, a compact reasoning model that outperforms similar-sized rivals like Gemini 2.5 Flash on math benchmarks—all while requiring a fraction of the compute.

🔍 Key Takeaways:

• Model Type: A distilled version of DeepSeek’s new R1, built on Alibaba’s Qwen3-8B (launched May 2025).

• Benchmark Wins: Surpasses Gemini 2.5 Flash on AIME 2025; nearly matches Microsoft’s Phi-4 on HMMT.

• Efficiency: Runs on a single 40–80GB GPU (e.g., Nvidia H100), compared to the full R1’s need for ~12x 80GB GPUs.

• Training Method: Fine-tuned on R1-generated outputs, emphasizing small-scale reasoning performance.

• Access: Released under the MIT license, with API access available via platforms like LM Studio.

💡 Why This Stands Out: Distilled models typically trade performance for efficiency, but DeepSeek’s latest 8B release narrows that gap—especially on high-difficulty math tasks. It underscores growing interest in models that balance capability with practical deployment needs. As labs chase bigger breakthroughs, will compact reasoning models become the real enablers of everyday AI?

Perplexity

Perplexity Launches Labs to Take AI Beyond Search

🚀 TodayOnAI Insight: Perplexity has launched Perplexity Labs, a new Pro-only workspace for creating reports, dashboards, code, and interactive apps—marking a major push beyond its core AI search engine and into productivity tools.

🔍 Key Takeaways:

• What’s New: Perplexity Labs lets Pro users generate structured outputs like spreadsheets, charts, and mini web apps using AI-powered workflows that run for 10+ minutes.

• Multiplatform Access: Available now on web, iOS, and Android—with Mac and Windows apps coming soon.

• Technical Capabilities: Labs integrates tools like live web search, code execution, and advanced file generation to handle complex research and analysis tasks.

• Organized Output: Generated files (charts, images, documents, etc.) are stored in a dedicated tab for download and review.

• Strategic Context: Launch follows Perplexity’s preview of its AI browser Comet and its acquisition of the professional network Read.vc.

💡 Why This Stands Out: Perplexity’s move into longer-form AI workflows signals a broader shift: search is becoming just one part of a more comprehensive AI productivity stack. As rivals like Manus debut similar tools, the race to own the AI workbench is heating up. Will users embrace these all-in-one AI suites—or prefer specialized agents?

✨ See how a simple prompt upgrade can unlock better AI output.

🔹 The Original Prompt

At first glance, this prompt might seem okay. But it's too broad — and that limits the quality of AI-generated results. Let’s improve it using prompt engineering best practices.

✅ The Improved Prompt

💡 Why It's Better

• Clarifies audience: Focuses on homeowners, not commercial clients or other contractors

• Adds strategic intent: Targets trust-building and lead generation

• Suggests useful themes: Budgeting, timelines, design — all pain points for customers

• Defines format: Bullet list ensures clean output

🛠️ Learn how to adapt this prompt for SaaS, AI tools, dev teams & more →
Read the full PromptPilot breakdown

💡 Bonus Tool: Want to generate and master prompts instantly?
👉 Try PromptPilot by TodayOnAI (Free to use)

📰 More from the AI World

• Buildots raises $45M to help companies track construction progress

• Odyssey’s new AI model streams 3D interactive worlds

• Meta AI now has 1B monthly active users

• Nvidia expects to lose billions in revenue due to H20 chip licensing requirements

🧰 Today’s AI Toolbox Pick

• 🎳Scenery (Video Design Tool): A collaborative video editing workspace for asset management, editing, and review.

• 💻ZipWP (Web Design Tool): Creates stunning websites in seconds.

• ⚡️Notion AI (Assistant Tool): Optimizes your notes, autofills tables, and answers questions.