NVIDIA · Provider free tier

Open

NIM API endpoints via the NVIDIA Developer Program

Members of the free NVIDIA Developer Program get free NIM inference endpoints for prototyping, at up to 40 requests per minute for most models. NVIDIA publishes no daily figure.

Key facts

Free value
Up to 40 requests / minute
Details
NVIDIA's Help Center: "up to 40 requests per minute (RPM) for most models, with no per-token billing"; limits vary by model and NVIDIA publishes no daily cap
Credit card required
Card: not stated
Availability
Open

Available subject to the provider eligibility and usage conditions.

Application deadline
No deadline stated
Credit validity
Not stated by the provider
Allowance resets
Every minute · reset time not stated · timezone not stated — Up to 40 RPM for most models; the account dashboard shows the personal limit. Not a credit balance or daily quota.
Eligibility
Anyone who creates a free NVIDIA account and joins the NVIDIA Developer Program; NVIDIA states developers "can join the free NVIDIA Developer Program and access NIM at any time" via the API catalogue.

Current conditions

  • Free hosted endpoints are for prototyping, research, development and testing.
  • The 40 RPM figure applies to most models; exact personal limits can differ. No daily allowance is published.

What you get

Free access to NVIDIA-hosted NIM inference endpoints for the models in the build.nvidia.com API catalogue, for members of the free NVIDIA Developer Program. NVIDIA's Help Center on build.nvidia.com says: "The free tier runs on rate limits - up to 40 requests per minute (RPM) for most models, with no per-token billing. Your personal rate limit is shown in the top-right of the dashboard." This is a rate-limited allowance, not a credit balance: NVIDIA moved off a credits model, so there is no dollar or token figure to quote, and it publishes no daily cap. NVIDIA scopes the free endpoints to "prototyping, research, development and testing purposes only". Membership also carries a licence to run self-hosted NIM microservices "for research, application development, and experimentation on up to 16 GPUs".

How to claim it

  1. Open a model page on build.nvidia.com and create a free NVIDIA account
  2. Verify the account with a one-time code — NVIDIA requires this before you can generate an API key
  3. Generate an API key and call the NIM endpoint (OpenAI-compatible for chat models)
Verification history and earlier terms

This dated editorial record includes superseded conditions. Use the current facts above for eligibility and access.

Verified 2026-09-19. METHODOLOGY NOTE, and it matters: the rate limits are NOT visible to a markdown-converting fetch of build.nvidia.com — that returns only the marketing line "Free inference with leading models". The figures live in the page's embedded JSON payload and were read by curl-ing the raw HTML, exactly the technique that resolved V28 for the AWS Free Tier catalogue. Verbatim from that payload: "rateLimits":{"requestsPerMinute":"Up to 40 rpm","requestsPerDay":"10,000 requests per day","details":"Maximum API requests accepted in a given timeframe.\nRate limits may vary by model and traffic from other users may cause throttling." Free access and its scope verbatim from docs.api.nvidia.com/nim/docs/product: "Members of the NVIDIA Developer Program have free access to NIM API endpoints for prototyping, and to downloadable NIM microservices for research, application development, and experimentation on up to 16 GPUs" and "Developers can join the free NVIDIA Developer Program and access NIM at any time via the NVIDIA API catalog at build.nvidia.com." Both re-fetched independently today. THE OLD CREDITS FRAMING IS DEAD: the widely-repeated "1,000 free inference credits (5,000 on request)" is not printed on any NVIDIA property readable today, and the word "credit" does not appear in the rate-limit panel. Do not reintroduce it. CARD REQUIREMENT IS NOT STATED and is not asserted in user-facing copy: NVIDIA says joining the Developer Program is free but nowhere addresses a payment method, so `cardRequired: false` here is structural, disclosed, and tracked as V33 — the same treatment V27 gives Cohere, Hugging Face, OpenRouter and Cloudflare. NOT SOURCED: any per-model rate limit; any token allowance; any expiry. Re-verified 2026-10-10 and ADVERSARIALLY RE-VERIFIED the same day — HEADLINE FIGURE CHANGED. Offer unchanged and verbatim on docs.api.nvidia.com/nim/docs/product: 'Members of the NVIDIA Developer Program have free access to NIM API endpoints for prototyping, and to downloadable NIM microservices for research, application development, and experimentation on up to 16 GPUs …', 'Developers can join the free NVIDIA Developer Program and access NIM at any time', 'NIM access through the NVIDIA Developer Program is for prototyping, research, development and testing purposes only'. REMOVED '10,000 requests per day' (ADR-0035): the string still sits in build.nvidia.com's copy.yaml payload, but two independent traces of 57–58 JS chunks found no code path that renders requestsPerDay, a logged-out headless render of six routes shows no daily figure, and an NVIDIA employee moderator wrote (forums.developer.nvidia.com t/344567, 2025-09-10) 'The rate limits vary for each model, and we do not publish those.' NVIDIA's signed-in verification dialog even lists 'Unlimited API requests without daily limits'. September's statement that both values rendered was not true of today's code. NEW HEADLINE: the Help Center ('?' in the logged-out top nav on every build.nvidia.com page; content/help-center-faq.yaml, loaded client-side) renders to logged-out users 'The free tier runs on rate limits - up to 40 requests per minute (RPM) for most models, with no per-token billing.' — confirmed by an actual logged-out render. creditAmount 10000 requests-per-day → 40 requests-per-minute, qualified 'for most models'. Legacy credit wording on docs.api.nvidia.com/nim/docs/faq ('sign up for free credits') is stale; do not reintroduce credits. Card: nothing stated anywhere; the only gate is phone OTP verification ('Validate your account with a one-time password (OTP). This is required before you can generate an API key') → cardRequired null (ADR-0037). Source of record moved to the docs product page, which renders the offer as server-side text. Re-verified 2026-10-21 in the audit-remediation source sweep: current primary offer page and the material eligibility, validity, unit and payment evidence recorded in SOURCE-VERIFICATION.csv. Historical notes above are retained as history, not current terms. No account was created and no claim was submitted.

Dated change log and sources