The Allen Institute for AI (AI2) dropped Olmo 3 on November 20, 2025 — a groundbreaking family of fully open-source large language models spanning 7B to 32B parameters, complete with every checkpoint, dataset, and training recipe from data curation to deployment. Featuring flagship reasoning beasts like Olmo 3-Think (32B) that match or beat Meta's Llama 3.1 and Alibaba's Qwen 3 on math, coding, and long-context tasks — all at 2.5x training efficiency — this release under Apache 2.0 license floods Hugging Face and AI2 Playground with tools for RL experiments and traceable outputs. It's not just models; it's the open-source revolution's full playbook, empowering devs to remix AI from the ground up without black-box mysteries.

Meta AI unveiled SAM 3D on November 19, 2025 — a groundbreaking extension of the Segment Anything Model family that reconstructs full 3D geometry, textures, and poses from just one everyday photo. Featuring dual powerhouses SAM 3D Objects (for scenes and clutter-crushing object meshes) and SAM 3D Body (for human shape estimation), it leverages a massive human-feedback data engine to outperform rivals on real-world benchmarks. With open checkpoints, inference code, and a new eval suite now live on GitHub and Hugging Face, SAM 3D slashes 3D capture time from hours to seconds — igniting a firestorm in AR/VR, robotics, and VFX where traditional multi-view setups just got obsoleted.

In a move that caught the AI community off guard, Elon Musk's xAI has quietly released Grok-4.1, the latest iteration of its flagship large language model. Unlike the fanfare surrounding its rivals' releases, this launch was subdued, yet its claimed capabilities are anything but. xAI asserts that Grok-4.1 demonstrates "state-of-the-art" performance across a comprehensive range of general-purpose benchmarks—from reasoning and coding to advanced mathematics—potentially surpassing established leaders like GPT-4o, Claude 3.5, and Gemini 2.0 in overall versatility and raw intellectual horsepower. This release marks a critical inflection point, challenging the perceived hierarchy of the frontier model race.

Google DeepMind has launched WeatherNext 2, its next-generation AI model that delivers hyper-accurate, high-resolution global weather forecasts up to 10 days in advance, within minutes. By integrating novel physics-informed neural architectures with a massive-scale multi-modal training approach, the model sets a new benchmark for predicting extreme weather events and fine-grained local phenomena, signaling a transformative shift from traditional numerical methods to agile, AI-native forecasting.

China's open-source AI community has achieved a major milestone: Zhipu AI's latest model, GLM-4.6, has ascended to a global #1 rank (tied) on the highly competitive LiveCodeBench (LCB) leaderboard. This achievement, surpassing giants like DeepSeek and closely trailing GPT-4, demonstrates a dramatic leap in practical programming capabilities for models outside the U.S. tech sphere, reshaping the global landscape of code-generation AI.

Chinese embodied AI startup Dexmal has secured nearly 10 billion RMB (approx. $1.4 billion USD) across two massive funding rounds led separately by Alibaba Group and NIO Capital. This colossal investment, one of the largest globally in AI for 2025, fuels the company's ambitious mission to develop a "one brain for all" general-purpose cognitive engine for robots and autonomous systems, signaling a fierce new battleground in the global race for advanced embodied intelligence.