Editorial standards

Selection

The directory covers model producers, research institutions, engineering labs and robotics companies. Its scope includes language, vision, speech, scientific AI, agents, inference systems, manipulation, locomotion and industrial automation. Selections are editorial, not a numbered league table of companies or universities.

Organizations and outputs

Each profile distinguishes an institution from the software, model family, dataset, robot or research program it produces. An inference engine is not a foundation model. A paper author is credited as a contributor; authorship alone does not establish current employment.

Backers

We record a funding relationship only with a source and a description of its scope. Philanthropy, government grants, research partnerships and equity investment are different relationships. A grant to an institute and a gift to its hosting laboratory are described separately.

Benchmarks

Scores stay within their original tests. The harness view uses SWE-bench Verified submissions; the local view uses Aider Polyglot results with a model-size filter; the frontier view follows LiveBench’s published release and suite weights. Dates, checked status and methodological qualifications remain visible. No composite score merges these incompatible measures.

Films and history

YouTube films remain on the original uploader’s channel and play through the uploader’s embed permissions. Archival footage is evidence of its period, not a claim that every forecast in a film came true. Timeline commentary distinguishes the original work from its later influence.

Country-level indicators

Stanford HAI’s Global AI Vibrancy Tool is a recurring source for national research, economy, talent, policy, public-opinion and infrastructure indicators. Its country scores are not lab rankings or model benchmark results. Global AI Map records selected organizations, outputs, papers, funding relationships and official-feed vacancies; it does not reproduce Stanford’s national totals or imply equivalent coverage. The current indicator mapping and known gaps are reviewed during monthly research.

Refresh policy

Public model and repository activity refreshes in the browser on visits and after 30 minutes while the page is visible. Benchmark pages fetch allowlisted public data and retain a dated static snapshot when a provider is unavailable. Institution profiles and historical accounts are pre-rendered and reviewed separately. Browser requests cannot rewrite the globally served HTML.

Review and commercial independence

Applications require verifiable identity, substantive work, primary evidence and clear limitations. Sponsor placement and affiliate income are disclosed and do not determine inclusion or ranking. A listing means the entry met our editorial criteria at review; it is not a product certification.

Corrections

Use the editorial application contact to supply a correction and its primary source. Every profile links to its supporting sources. Map pins are approximate and do not identify exact facilities unless an address is explicitly provided.