Google pushed the Gemini line forward in 2025 with the Gemini 3 family — a multimodal, higher-reasoning model (Gemini 3 Pro and Deep Think modes were called out in Google’s announcement). The rollout emphasized multimodal reasoning across text, images, audio and video and tighter integration into Google products.
“Manus” (meaning “hand”) is an example of the new wave of autonomous AI agents: systems designed not only to answer questions but to plan, execute and iterate on multi-step tasks with minimal human oversight. The product pages and contemporaneous reporting describe the agentic approach and its practical uses.
Video generation moved from demos to production-focused tools in 2025. Lightricks’ LTX (and the LTX-2 model) announced higher-fidelity, synchronized audio+video pipelines and open tooling for creators — a big step for content creators and short-form filmmaking workflows.
SmolVLA showed it’s possible to build capable vision-language-action models that fit on consumer hardware. The project published a paper and released community code/models aimed at lowering the compute barrier for robotics research and hobbyists.
Helix demonstrated a hybrid “System-1/System-2” VLA approach for high-rate, dexterous control of humanoid upper bodies — a notable step toward robots that can perform coordinated, fine-grained manipulation in the real world.
Multiple labs published work showing AI can accelerate climate and Earth-system simulations — producing longer runs and higher local resolution faster than some traditional approaches. This has big implications for forecasting and policy, though researchers emphasize careful validation.
In 2025 researchers and companies continued integrating large multimodal models into robots, drones and other “embodied” agents. The effect: better perception, planning and closed-loop control in real environments — but also renewed focus on safety and reproducibility.
2025 saw broader geographic and organizational diversity in important AI projects — open models, university groups, startups and regional players produced influential work, shifting the innovation map away from a tiny handful of companies.
AI’s role in diagnostics, drug discovery, scientific simulation and critical infrastructure continued to grow — with pilot deployments and stronger regulatory attention. The headline: more impactful, high-stakes uses and therefore more scrutiny.
As capability raced forward, policymakers and researchers pushed harder on safety, transparency, and affordability — from independent audits to new open models that make advanced capabilities more widely available while stressing guardrails.