AI tools for AI Infrastructure

Atoms.dev is a multi‑agent product development platform that transforms plain‑language ideas into fully functional, production‑ready web and mobile applications, without requiri

DeepInfer is DeepSeek's inference framework for efficient, large-scale execution of supported DeepSeek models. It combines optimized kernels, expert parallelism, asynchronous sc

Unpaired Rosetta is a cross-modal embedding-alignment implementation that does not require paired examples. Wasserstein Procrustes alignment estimates an orthogonal map between

OpenDots is CopilotKit's open application for AI coworkers that stay available across text conversations, calls, and Slack. The application connects persistent agent work and co

PDMD, or Projected Distribution Matching Distillation, is a method and implementation for faster diffusion generators. A projection removes the component of the training directi

AstaBrief is Ai2's 8B open-weight model for producing cited scientific reports from questions and retrieved literature. It synthesizes supplied research excerpts into structured

DMAD is an adversarial distillation framework for reducing the sampling steps of visual generators. Two discriminator heads on a shared backbone classify teacher, student, and r

Olmo-core 3 is Ai2's open training infrastructure for large language models and mixture-of-experts systems. A redesigned MoE training stack distributes model development across

DeepGEMM Ascend is DeepSeek's efficient matrix-operation library for Huawei Ascend hardware. It exposes DeepGEMM-compatible APIs over Ascend matrix primitives while handling pla

OpenMuse is a self-hostable personal-agent application with browser, terminal, and file access across mobile and web interfaces. Built with CopilotKit and AG-UI, it connects an

Aikido Altar is an open-weight security model built for defensive intelligence inside controlled infrastructure. It compresses a GLM-5.3-derived model with quantization and expe

Hy4 Preview is Tencent mixture-of-experts model for coding, professional analysis, game development, and scientific work. It emphasizes extended productivity workflows over larg

LLaDA-Image is a six-billion-parameter model family for image generation and instruction-guided editing. A single checkpoint can create an image from text or alter a reference w

SkySynth, also called SkyDiscover-Synthesize, uses coding agents to build systems specialized for a known workload and deployment environment. It targets situations where a gene

OpenWAM is a modular research stack for studying how video world models can learn executable robot control. It separates infrastructure, controlled design experiments, and the r

DeepSeek-V4.1-Flash is a multimodal mixture-of-experts model for text reasoning, image understanding, and agent workflows over very long inputs. It supports contexts of up to on

Qwen3.8-2.4T-A95B is a very large mixture-of-experts language model with 2.4 trillion total parameters and 95 billion activated parameters. It targets advanced coding, agentic w

MatrAIx is simulated-user evaluation infrastructure for digital products and AI systems. It models a population of 8.3 billion persona agents so teams can test how different use

Ideogram 4 is an open-weight image generation model built for design work, realistic 2K outputs, multilingual text rendering, and precise layout control. It is designed for crea

SANA-Video 2.0 is an NVIDIA research project for efficient video generation with hybrid linear attention and attention residuals. It targets high-quality video up to 720p while

Agents-A1 is a 35B mixture-of-experts agentic language model designed for long-horizon search, engineering, scientific research, instruction following, and tool calling. It targ