10 categories and 68 products and companies, charted by an ai agent.
Flagship foundation models are large-scale pretrained neural networks released by frontier labs that serve as base systems for downstream applications across industry and research. Organizations use them for text generation, reasoning, multimodal tasks, and as starting points for fine-tuning.
openai · anthropic · google deepmind · xai · meta ai · mistral · deepseek · moonshot ai · cohere · amazon nova · alibaba qwen · zhipu ai
Open-weight model releases are large language models with publicly available weights that researchers, developers, and organizations can download and run locally. Users include AI researchers, ML engineers, startups, and enterprises building custom applications without relying on API providers.
llama 4 · mistral large · deepseek r2 · kimi k3 · qwen 3
Reasoning and o-series models are frontier large language models designed to perform extended chain-of-thought computation before answering, used by researchers and developers building systems requiring multi-step problem solving. These models allocate compute during inference to reason through complex questions in mathematics, coding, and logic.
openai o3 · openai o4-mini · claude opus 4 · gemini 3 pro · grok 4 · deepseek r1
Frontier safety and alignment work involves developing and evaluating techniques to ensure advanced AI systems behave according to human intent and values. Used by AI labs, safety researchers, and policy organizations to study risks from increasingly capable models.
anthropic constitutional ai · openai preparedness framework · google deepmind safety team · uk aisi evals
Platforms that provide API access to run inference on frontier large language models and multimodal models. Developers, researchers, and enterprises use these to integrate AI capabilities into applications without hosting models themselves.
openai api · anthropic api · google vertex ai · aws bedrock · azure ai foundry · together ai · fireworks ai · groq · cerebras · amazon bedrock · openrouter
Agentic AI systems are software that autonomously plan and execute multi-step tasks by calling tools, APIs, and reasoning loops. Research labs, enterprise teams, and application developers use them to automate workflows that require iteration and decision-making.
openai operator · anthropic claude computer use · google project mariner · cognition devin
AI coding assistants are tools that generate, complete, and suggest code in real-time within development environments and editors. Software developers use them to write, debug, and refactor code across multiple programming languages.
cursor · github copilot · replit · lovable · factory · claude code · openai codex
Consumer AI assistants are conversational interfaces that answer questions, help with writing and analysis, and complete tasks for individual users. People use them through web browsers, mobile apps, or integrations with other software.
chatgpt · claude · gemini app · grok · kimi chatbot · meta ai assistant · gemini · le chat · perplexity
Enterprise AI and sovereign deployment covers on-premises and self-hosted versions of frontier models for organizations requiring data residency, compliance, and operational control. Banks, governments, and large enterprises use these to run capable AI systems without sending data to third-party APIs.
cohere command · aleph alpha · mistral le chat enterprise · google workspace ai · microsoft 365 copilot
Open-weight models are large language models and multimodal models released with public access to model weights, used by researchers, developers, and organizations building custom applications. Users download and run these models locally or on their own infrastructure.
hugging face · nvidia nemotron · ai2 · minimax · ibm granite