Skip to search
BigHugger
Request accessSign in

There’s already a model for that.

4.4 million open models, datasets, papers, repositories, packages, MCP servers and skills, crawled — the ones worth returning, searched as one index.

Plain words are fine. Enter to search.

Every model, dataset and paper in AI, indexed deeply enough to answer questions about it.

  1. One index for the whole ecosystem

    Models, datasets, papers, repositories, packages, MCP servers, skills and frameworks. Eight catalogues that have never spoken to each other, resolved into one thing you can query in a single call.

    4.4Mentities indexed

  2. Indexed down to the file

    Parameter count, quantisation, context length, licence class — the facts that decide whether a model actually fits live inside config, tokenizer and chat-template files. We read them, so you can query on them.

    4.7Mfiles read

  3. Structured, so you can compare

    Every property is a field you can filter on and every benchmark a number you can sort by, attached to the thing it describes. Choosing between two models stops being a research project and becomes a query.

    7.1Mstructured facts

  4. Ask it in plain language

    One endpoint searches the index and one agent reasons over it, working from the ranked and enriched core of the corpus. Every id, size, format and licence in an answer is looked up as the answer is written, so what comes back is checkable.

    3.4Mranked for retrieval

6.4 quadrillion parameters measured across 874,280 models. Read live from the index, refreshed every six hours. 2026-09-18.

Products

Search~0.8s$0.0001 a call

Vector and keyword retrieval across all eight corpora, fused and ranked, returned as rows with the facts each kind is judged by. No model reads the results and none writes anything. The search box on this page runs exactly this.

Ask~40s$0.05 a call

The agent. Plans the search, runs it across the index and the open web, reads what comes back, and returns one recommendation with its reasoning and its sources — every model id, parameter count, file size and licence checked against the index rather than recalled. Streams newline-delimited JSON, one self-contained object per line.

Workbooks~5s$0.0083 a call

Any answer, packed as something that runs: a notebook, a pinned environment, the exact model file, an MCP server and a SKILL.md. Workbooks outlive the conversation that made them and stay pullable long afterwards. Priced on top of the ask whose answer it packs.

Model cacheasync$0.018 per GB-month

Mirror a repository's weights to our storage so a machine that needs them twice does not fetch them twice, and a pinned revision stays pinned however the upstream moves. Pulls are served from the edge and support range requests, so a resumed download resumes.

Times and prices are measured. Copy any product and your agent gets the command, the route and the docs link as one block.

Pricing

Pay for the calls you make, from a balance you top up. No subscription, nothing expires, and every account starts with $5 of tokens to find out whether it is useful.

  • Rate$1per 10,000 tokensOne unit for every callSearch 1 · Ask ≈520 · Workbook +83
  • Top up$20minimum200,000 tokens ≈ 384 answersAny amount above it, by card. The balance never expires.
  • Model cache$0.018per GB-monthUp to 500 GB per accountDrawn daily from the same balance for the weights you hold.

Every account gets the full index, workbooks, the MCP server and API keys. Your agent sees its balance and a top-up link when it runs low.

Request access

Access is invite-only. Ask, and we will get you a key.

Request access