AI models catalog
Epoch AI notable-models catalog (CC-BY 4.0). Each model carries its developer, domain, parameters, training compute (FLOP), training hardware, frontier flag, and accessibility. Rows can be sorted by publication date or training compute. Requires a plan with AI compute & models access (can_use_epoch_ai) — the Corporate & Supply Chain and Global Intelligence plans.
Boolean fields (frontier_model, open_weights) are real JSON booleans, training_compute_flop is de-artifacted, and the response echoes applied_filters (unknown params surface in applied_filters.ignored). This product uses data from Epoch AI (CC-BY 4.0).
Authorizations
GDELT Cloud API key. Send as Authorization: Bearer gdelt_sk_....
Query Parameters
Case-insensitive substring match on the developing organization (e.g. OpenAI, Google, Anthropic, xAI, Meta AI). Alias org is accepted.
Case-insensitive substring match on task domain. The enum lists the atomic domain vocabulary (models can span several, e.g. Vision,Multimodal — so Vision also catches those rows); any substring is accepted.
Language, Vision, Biology, Multimodal, Image generation, Video, Speech, Robotics, Games, Audio, Medicine, Mathematics, Materials science, Earth science, Recommendation, Driving, Search, 3D modeling, Other Release/access tier — exact, case-insensitive match on the stored value. Use the open-weights tiers to filter to models with released weights.
Open weights (unrestricted), Open weights (restricted use), Open weights (non-commercial), API access, Hosted access (no API), Unreleased Case-insensitive substring match on the developing organization's category. Accepted tokens: Industry, Academia, Government, Research collective. A model may carry several comma-composited categories, so a token matches any model with that contributor type (e.g. Academia = any model with an academic contributor, not academia-exclusive). Not strictly validated — an unmatched value returns 0 rows rather than 400.
true → only models Epoch flags as frontier; false → the non-frontier rows. Accepts true/false/1/0.
Floor on training_compute_flop (e.g. 1e25).
Filter to models with publication_date on or after this date (YYYY-MM-DD). A malformed date is ignored rather than erroring.
"2024-01-01"
Filter to models with publication_date on or before this date (YYYY-MM-DD). Combine with date_start for a publication-date window.
"2024-12-31"
Sort by publication date (default) or training compute (FLOP), both descending.
publication_date, compute Rows per page (1–200). Defaults to 25.
1 <= x <= 200Canonical page token — pass the pagination.next_cursor from the previous response here to fetch the next page (uniform across all list endpoints).
Row offset for pagination. Prefer the canonical cursor token; both are supported.
x >= 0