Replicate
Hosted API and cloud execution for Replicate and third-party models; each model’s license and provenance remain distinct.
Score calculation
(6 + 4) ÷ 2 = 5.0
2 of six dimensions rated. Unrated dimensions and BDS status are excluded.
Full scoring method ↗Setup details
Replicate operates a model marketplace/API; models may be third-party offerings, and associated provider terms govern some rights.
Energy
↑Replicate's terms describe managed model runs and configurable deployments, including customer choices of hardware and scaling, but do not identify serving-facility energy, emissions, or water. Hardware choice is useful context, not a usage-linked footprint.
Reasoning & sources
Replicate lets customers deploy models with hardware and scaling choices, but its public terms and model catalog do not provide measured power or water for an actual run. Request region-specific, workload-linked footprint data for the selected model and deployment; do not substitute GPU type or runtime for measurement.
Model-specific hardware and deployment locations are not tied to measured per-request electricity, emissions, or water in the cited terms/docs.
Training rights
↑Terms define Marketplace Models as third-party offerings and point to separate Third Party Terms; service terms do not establish training corpus permissions for every model.
Reasoning & sources
Terms define Marketplace Models as third-party offerings and point to separate Third Party Terms; service terms do not establish training corpus permissions for every model. Select the model first and review its license plus corpus-provenance disclosures; prompt no-training terms do not clear the model’s original dataset.
Model-by-model license and dataset provenance require review.
Privacy
↑Replicate’s API documentation says prediction inputs, outputs, files, and logs are automatically removed after one hour by default; predictions made through the web interface are kept indefinitely. Its general privacy policy permits retention as needed for service, legal, abuse-prevention, and claims purposes and does not establish a blanket no-training promise for every hosted model.
Reasoning & sources
A mid-range privacy judgment is supported for API predictions because content is removed after an hour by default, while web predictions persist and the general policy preserves service, legal, and abuse-prevention retention. Use the API only after checking the model and feature terms; do not assume the one-hour rule applies to web use or establishes a no-training guarantee.
The API has an explicit one-hour default cleanup for prediction content, but web predictions are retained indefinitely and the general privacy policy does not establish a universal model-training prohibition.
Confirm whether the user is calling the API or web interface, whether the model or feature has additional handling, and whether a customer contract changes backup, security, or legal-retention periods.
Openness
↑Replicate provides a stable API and model marketplace, but each model has separate terms and the hosted service does not make its runtime portable.
Reasoning & sources
The catalog enables model choice, but hosting and portability vary by model and the service remains managed. Check whether the chosen model’s weights and license permit inspection or relocation, and whether the serving runtime moves too; API compatibility alone is not an open stack.
The catalog enables model choice, but hosting and portability vary by model and the service remains managed.
No claim can be made from silence; additional direct documentation or independently verified records are needed.
Worker treatment
↑Replicate's terms distinguish the platform from Marketplace Models and third-party model terms; its official-model directory shows that hosted models originate with external model developers. These sources do not report worker conditions for those developers, Replicate contractors, or data labeling.
Reasoning & sources
Replicate's model directory and terms show that the hosted catalog includes third-party models with separate terms, so model-development labor cannot be attributed to Replicate or assumed to be covered by its API. The reviewed sources give no worker outcomes for either platform contractors or model teams; request disclosures from both the host and selected developer.
No current provider-specific worker pay, location, exposure safeguards, audit, or remediation findings were established for the hosted catalog.
Government & military
↑Replicate's current terms prohibit certain weapon-related and high-risk uses, including weapons-of-war development and weapons-system applications. This is a platform use restriction; the reviewed sources do not identify a defense contract or military deployment for Replicate's inference service.
Reasoning & sources
Replicate's terms restrict weapons-of-war and specified high-risk uses, which is a relevant contractual safeguard, but they do not document actual defense contracts or prove enforcement in a particular deployment. Treat actual deployments as unverified and request any named government award, service scope, and compliance evidence.
Terms do not establish how restrictions are enforced in a particular deployment or rule out undisclosed public-sector customers.
BDS status
↑We did not find Replicate, LLC among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.
Reasoning & sources
Replicate, LLC is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Your selected model, cloud and billing route can introduce other companies into the purchase.
This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.