AI Research & Analytics
-
NVIDIA Cosmos Reason 2 is an open vision-language model built to reason about the physical world
NVIDIA Cosmos Reason 2 is an open reasoning vision-language model with a narrow but hard job: let machines see a physical scene, understand what’s happening, and decide how to act. It’s the reasoning brain in NVIDIA’s broader Cosmos physical-AI stack. ## Reasoning, not just captioning Most vision-language models describe an image. Cosmos Reason 2 is… Continue reading
-
Leni is a vertical AI analyst that beat OpenAI and Google on deep-research benchmarks for finance
Leni is a purpose-built AI analyst for commercial real estate and private equity — not a general chatbot pointed at finance, but an agentic system that handles underwriting, portfolio reporting, market research, and memo generation off data pulled from property and finance systems. ## The benchmark claim What got it attention is accuracy. Leni placed… Continue reading
-
NVIDIA’s GRAIL generates humanoid robot skills from 3D assets and video
Teaching a humanoid to pick up a box and carry it across a room usually means hours of teleoperation or motion capture. NVIDIA’s GRAIL goes the other way: it generates loco-manipulation behavior — locomotion plus object handling — directly from 3D assets and motion priors pulled from video. ## The data bottleneck it targets Humanoid… Continue reading
-
NVIDIA Cosmos 3 is an open world model that generates text, video, sound, and actions
## What it is NVIDIA just released Cosmos 3, an open foundation model for Physical AI that natively understands and generates across text, images, video, ambient sound, and actions — all in one model. It’s built on a two-tower Mixture-of-Transformers: an autoregressive transformer handles physical reasoning while a diffusion transformer handles multimodal generation. The point… Continue reading
-
Foundation Protocol proposes a coordination layer for an ‘agentic society’ of many AI agents
Foundation Protocol is a proposed coordination layer for what its authors call an “agentic society” — the emerging world where many AI agents, built by different parties, need to discover, negotiate with, and coordinate with each other. ## Where it fits in the protocol stack 2026’s agent infrastructure has been assembling in layers: MCP (agents… Continue reading
-
Macaron-A2UI is a model for generative UI in personal agents — interfaces the agent builds on the fly
Macaron-A2UI is a model for generative UI in personal agents: instead of returning text, the agent generates an actual interface — buttons, forms, charts — tailored to the task at hand. It builds on A2UI, the framework-agnostic generative-UI protocol that lets agents render components across web, mobile, and desktop. ## The generative-UI shift 2026 has… Continue reading
-
Rixx is a Perplexity alternative built around the whole research workflow, not just the answer
Rixx positions itself against Perplexity with a different emphasis: where Perplexity optimizes for delivering a cited answer, Rixx is built around the entire research process — search, explore, analyze, and turn findings into structured outputs, all in one connected workflow. Its tagline: “from search to publishable research in one workflow.” ## What’s different Both give… Continue reading
-
SciAtlas builds a large-scale knowledge graph to automate scientific research
SciAtlas, from UCL, is a large-scale knowledge graph aimed at automating scientific research — structuring the relationships across papers, methods, datasets, and findings so an AI system can navigate the literature the way a domain expert does. ## The problem Scientific literature expands faster than any human can track. Automated knowledge-graph construction is a hot… Continue reading
-
SkillOpt proposes an executive strategy for agents that decide which skills to evolve and keep
SkillOpt, from Microsoft Research, is the top-upvoted agent paper on Hugging Face today. It tackles a layer above individual skills: the executive strategy for a self-evolving agent — deciding which skills to create, refine, keep, or discard as the agent accumulates experience. ## The self-evolving-skills space A growing body of work frames the same closed… Continue reading
-
WorldKV lets world models remember what they have seen — training-free KV retrieval and compression for spatial consistency
WorldKV, from KAIST AI and Naver AI Lab, tackles a core problem in world models: when you revisit a place you’ve already seen, the model should show you the same thing. Sustaining that persistent consistency has been hard — full attention preserves it but blows the real-time budget; sliding-window inference is fast but forgets. ##… Continue reading
Top AI Product 