NASA Launches Its First-Ever AI Model

NASA & IBM’s Surya AI forecasts solar storms, ByteDance’s Seed-OSS-36B advances long-context reasoning, and MindJourney gives AI 3D spatial exploration skills.

This week’s breakthroughs highlight how AI is pushing the boundaries of science and perception. NASA and IBM’s Surya AI is transforming space weather forecasting by predicting solar flares hours before they strike, while ByteDance’s Seed-OSS-36B sets a new standard for long-context reasoning with open-source accessibility. At the same time, MindJourney is giving AI the ability to explore and interpret 3D environments, moving beyond flat images to dynamic spatial understanding. Together, these innovations show how AI is becoming more predictive, more adaptable, and more capable of decoding the complexities of our world and beyond.

Solar Storms Decoded: NASA & IBM’s Surya AI Predicts Space Weather

NASA and IBM have unveiled Surya, a pioneering AI foundation model trained on nine years of high-resolution data from NASA’s Solar Dynamics Observatory, which analyzes solar behavior to forecast solar flares and predict space weather threats to satellites, power grids, and communications. Surya outperforms previous benchmarks by 16%, generating visual predictions of solar flare activity two hours ahead, a significant advance for operational space weather forecasting. Utilizing continuous data coverage of an entire solar cycle, Surya adapts rapidly to diverse scientific tasks without the need for extensive data labeling and offers open access via HuggingFace and GitHub, empowering researchers worldwide to build upon its versatile architecture for heliophysics, Earth observation, planetary science, and beyond. This project, supported by the National Artificial Intelligence Research Resource Pilot and developed collaboratively by NASA, IBM, NSF, and leading universities, lays the groundwork for future AI-driven tools to decode complex phenomena affecting modern technology and society.

Advanced AI with Long-Context and Flexible Reasoning

Seed-OSS-36B-Instruct is a cutting-edge, open-source large language model developed by ByteDance’s Seed Team, designed for advanced text generation, reasoning, and agentic intelligence with versatile developer-friendly features and support for long-context (up to 512K tokens). Trained on 12 trillion tokens, the model excels on popular benchmarks across knowledge, reasoning, math, coding, and multilingual tasks while allowing users to flexibly control the "thinking budget" for tailored inference performance. Released under the Apache-2.0 license for research and commercial use, Seed-OSS includes pre-trained versions both with and without synthetic instruction data, enabling broader experimentation. Built for international applications and optimized for research, Seed-OSS-36B-Instruct is available on Hugging Face, supports efficient inference via Transformers and vLLM, and comes with extensive documentation, community support, and scalable quantization options—all reflecting ByteDance Seed Team’s commitment since 2023 to advancing AI for science and society.

MindJourney Empowers AI with 3D Spatial Exploration and Interpretation

MindJourney is a novel AI research framework that enhances vision-language models (VLMs) by enabling them to explore and interpret simulated 3D environments, addressing their limitations in spatial reasoning from static 2D images. Inspired by how humans mentally navigate spaces, MindJourney uses a world model trained on moving video perspectives to generate photorealistic views from different viewpoints, guiding AI agents through an efficient spatial beam search that prioritizes the most informative paths. This iterative process of simulating, evaluating, and integrating new observations boosts VLM accuracy by 8% on spatial reasoning benchmarks without additional training. By combining symbolic VLM capabilities with learned spatial patterns from world models, MindJourney brings AI closer to understanding his physical world dynamically, with potential applications in autonomous robotics, smart homes, and accessibility through better environment interpretation and planning. Future extensions aim to add predictive modeling of scene changes over time to further improve AI agents’ adaptability and decision-making in dynamic real-world settings.

Hand Picked Video

Create privacy-first videos with AI that blur faces, number plates, and backgrounds in real time to protect sensitive information while maintaining video clarity. Upload your video for instant privacy enhancement.

Top AI Products from this week

  • ClearoBraindump anything. Clearo processes and organizes your raw input into actionable items (tasks, reminders, goals, events..) and assigns due dates, times, and repeat settings. Chat with Clearo to create entries or manage your powerful calendar and kanban board.

  • WriteMindAI WriteMind AI is your intelligent writing companion that converts scattered web research into well-structured, citation-ready articles. Perfect for creators, students, journalists, and professionals who want to save time and boost quality.

  • Crawl4Empower your research or products through our Playground or API with clean, structured data from any website.

  • AskaiWPSummarize, explain, and chat — directly on your WordPress site.

  • Nivy Commerce is fragmented; your life isn't. Nivy is the single intelligent layer to manage it all. An AI agent that lets you discover, buy, and automate purchases from anywhere, just by asking. This is your personal portal to all the commerce, product or service.

  • Farmers AssistantFarmer's Assistant is an innovative chat bot designed specifically for farmers. The platform leverages the latest AI to provide a user-friendly experience that allows farmers to access information and assistance instantly in a simplified manner via WhatsApp.

This week in AI

  • The Meta Ray-Ban Glasses Rival - Harvard dropouts launch Halo X, AI glasses that record, transcribe conversations, display real-time info, raising privacy concerns over covert recording.Heading: AI Smart Glasses

  • Claude Code Boost - Claude Code adds premium coding AI to team plans, enabling seamless dev workflows, admin controls, spend caps, and Compliance API for real-time usage and governance data.

  • Z.ai & Alibaba AI Agent - Chinese AI unicorn Z.ai and Alibaba Cloud launch free AI agent on Android/iOS to handle tasks like booking, ordering, and content creation via natural language commands.

  • Character.AI’s Future - Character.AI weighs sale or funding at $1B+ valuation amid rising costs, lawsuits, and regulatory scrutiny while growing revenue with 20M users and new features.

  • Grammarly AI Redesign - Grammarly unveils a new document-based interface with AI tools including assistant, grader, citation finder, paraphraser, and AI content detectors to aid writing and learning.

Paper of The Day

Google’s Gemini AI assistant uses only 0.24 Wh of energy, emits 0.03 grams of CO2, and consumes 0.26 mL of water per text prompt—less energy than watching TV for nine seconds. Efficiency improvements over a year have cut energy use 33x and emissions 44x, setting a new standard for measuring AI’s environmental impact at scale.

To read the whole paper, go to here.