The open knowledge graph of AI models.
What each model can do, what it costs, where it runs, and how it scores on every benchmark we can find. Free, open source, and built for anyone who has to choose a model and be right.
01The graph
Every model as a node, linked to its provider, its lineage, its competitors, the platforms that serve it, the benchmarks that measured it, its capabilities and its licence. Eight views, one graph, explorable in 3D.
02The downselect
Tell it your use case, where you can host, what hardware you have, from a Raspberry Pi to a B200, and the rules you live under: open weights, origin, cost, context. It filters, scores, ranks, and shows its reasons.
03The CLI
The same graph from your terminal, for scripts and for the tools that build with you.
modelspec rank --use-case coding --top 10 modelspec compare anthropic/claude-opus-4-6 openai/o3 modelspec hardware macbook_air_m4_24gb modelspec gaps --top 20
The site is coming online. The data, the graph and the CLI are open source on GitHub today. Next: research on every model and every benchmark, every day, merged by a human; a graph of the benchmarks themselves; and an authoring guide for each model, so you know how to talk to it before you spend a token.
ModelSpec is built and kept current by the team behind Dark Product Factories, a guided factory that walks a product from a customer problem to production. ModelSpec is the model axis of that factory, and it is free for everyone.