NVIDIA Build / NVIDIA-hosted NIM
No-cost research and development inference across 82 models; not unrestricted production use.
https://integrate.api.nvidia.com/v1Models mentioned
01-ai/yi-largeadept/fuyu-8bai21labs/jamba-1.5-large-instructaisingapore/sea-lion-7b-instructbigcode/starcoder2-15bdatabricks/dbrx-instructdeepseek-ai/deepseek-coder-6.7b-instructdeepseek-ai/deepseek-v4-flash-0731z-ai/glm-5.3google/codegemma-1.1-7bgoogle/codegemma-7bgoogle/deplotgoogle/diffusiongemma-26b-a4b-itgoogle/gemma-2bgoogle/gemma-3-12b-itgoogle/gemma-3-4b-itgoogle/gemma-4-31b-itgoogle/recurrentgemma-2bibm/granite-3.0-3b-a800m-instructibm/granite-3.0-8b-instructibm/granite-34b-code-instructibm/granite-8b-code-instructmeta/codellama-70bmeta/llama-3.2-11b-vision-instructmeta/llama-3.2-90b-vision-instructmeta/llama-guard-4-12bmeta/llama2-70bmeta/muse-glimmer-30bmicrosoft/kosmos-2microsoft/phi-3-vision-128k-instructmicrosoft/phi-3.5-moe-instructmistralai/codestral-22b-instruct-v0.1mistralai/mistral-7b-instruct-v0.3mistralai/mistral-largemistralai/mistral-large-2-instructmistralai/mistral-nemotronmistralai/mixtral-8x22b-v0.1moonshotai/kimi-k2.6moonshotai/kimi-k3nv-mistralai/mistral-nemo-12b-instructnvidia/ai-synthetic-video-detectornvidia/cosmos-reason2-8bnvidia/embed-qa-4nvidia/ising-calibration-1.5-31bnvidia/llama-3.1-nemoguard-8b-content-safetynvidia/llama-3.1-nemoguard-8b-topic-controlnvidia/llama-3.1-nemotron-51b-instructnvidia/llama-3.1-nemotron-70b-instructnvidia/llama-3.1-nemotron-safety-guard-8b-v3nvidia/llama-3.1-nemotron-ultra-253b-v1nvidia/llama-3.2-nemoretriever-1b-vlm-embed-v1nvidia/llama-3.2-nv-embedqa-1b-v1nvidia/llama-nemotron-embed-vl-1b-v2nvidia/llama3-chatqa-1.5-70bnvidia/mistral-nemo-minitron-8b-8k-instructnvidia/nemotron-3-embed-1bnvidia/nemotron-3-nano-omni-30b-a3b-reasoningnvidia/nemotron-3-super-120b-a12bnvidia/nemotron-3-ultra-550b-a55bnvidia/nemotron-3.5-content-safetynvidia/nemotron-3.5-lightning-30b-a3bnvidia/nemotron-4-340b-instructnvidia/nemotron-4-340b-rewardnvidia/nemotron-nano-3-30b-a3bnvidia/nemotron-parsenvidia/nemotron-parse-2.0nvidia/neva-22bnvidia/nv-embedqa-mistral-7b-v2nvidia/nvclipnvidia/riva-translate-4b-instructnvidia/riva-translate-4b-instruct-v1.1nvidia/riva-translate-4b-instruct-v2nvidia/vilaopenai/gpt-oss-20bpoolside/laguna-xs-2.1snowflake/arctic-embed-lwriter/palmyra-creative-122bwriter/palmyra-fin-70b-32kwriter/palmyra-med-70bwriter/palmyra-med-70b-32kShowing 80 of 82. Use the provider discovery URL for the complete live catalog.
Equivalent paid value
How this was valued: Best-case rate-limit envelope for one current NVIDIA free endpoint, moonshotai/kimi-k3. NVIDIA's live model page advertises a free endpoint, a 1,048,576-token context, up to 40 requests/minute, and 10,000 requests/day; its API reference caps output at 65,536 tokens/request. NVIDIA sells no token-priced hosted tier, so the exact-model comparison is OpenRouter's current moonshotai/kimi-k3 route at $2.648138063 input / $13.28272425 output per 1M tokens (snapshot 2026-09-13). The result assumes uninterrupted maximal-context saturation and is one model's ceiling, not an additive total across the 82-model NVIDIA catalog.
Partial because NVIDIA's own limit text says rates "may vary by model and traffic from other users may cause throttling", the phrasing is "up to", the allowance is development/prototyping-only, and the paid reference is a router price rather than an NVIDIA rate for the same service.
Limits and terms
- Public Numeric Limits
- No
- Official Rule
- Varies by model and concurrency; inspect the signed-in account.
- Current Marketing Phrase
- Unlimited prototyping
- Allowed Use
- prototypingresearchdevelopmenttestinglearning
What happens to your prompts?
The free developer terms permit storage and broad product, service, and underlying-model improvement uses.
- Plan Scope
- NVIDIA API Catalog and NVIDIA-hosted NIM trial access under the Technology Access Terms; a model or product-specific agreement can supersede those terms.
- Prompt Retention
- No fixed prompt-retention period is promised. The terms permit NVIDIA, affiliates, and service providers to host and store User Content to provide and support the service, for security, and to improve products or underlying technology.
- Response Retention
- No inference-output-specific retention commitment was found in the governing trial terms.
- Ordinary Logging
- NVIDIA may monitor, scan, or review communications and User Content transmitted through its servers for safety, security, moderation, or legal requests.
- Model Training
- The terms do not make a narrow no-training commitment. Their User Content license permits modification and improvement of NVIDIA products, services, and underlying technology, so users should treat training or equivalent improvement use as permitted unless a product-specific agreement says otherwise.
- Product Improvement
- Expressly permitted under the User Content license.
- Human Or Operator Access
- Monitoring, scanning, and review are permitted for security, moderation, and legal purposes; service providers may process content under the license.
- Subprocessors And Routing
- NVIDIA affiliates and service providers may process User Content, and separately identified product agreements can apply to individual models or services.
- Deletion Controls
- The terms do not guarantee permanent access or recovery if data is deleted or lost; public User Content may remain after reposting, and NVIDIA accepts no general duty to remove it.
- Caveat
- Unless a separate product agreement expressly permits it, User Content must not contain confidential information, personal data, protected health information, payment-card data, or sensitive human-subject research. This makes the free trial unsuitable for sensitive prompts.
Governing documents
- Technology Access Terms
- https://assets.ngc.nvidia.com/products/api-catalog/legal/NVIDIA_Technology_Access_TOU.pdf
Eligibility
- Account Required
- Yes
- Developer Program Membership Required
- Yes
- Membership Cost USD
- 0
- Payment Method Required
- not documented
Primary sources
Before you build with NVIDIA Build / NVIDIA-hosted NIM
Read the classification narrowly
This record has current first-party evidence for some zero-cost hosted inference. The label describes the bounded offer supported by evidence on 2026-09-19; it does not guarantee permanence, production suitability, uptime, latency, model quality, or access from every region.
Resolve the live model route
This snapshot records 82 model IDs. Match the exact ID against the provider's live catalog before using it in code; zero-price routes and aliases can rotate while an older documentation page remains online. The recorded base endpoint is https://integrate.api.nvidia.com/v1, but the provider's current API reference remains authoritative for paths, authentication, and request shape.
Confirm account and billing boundaries
An account is required, so public catalog evidence cannot by itself prove the limits, balance, or billing behavior attached to your workspace. The reviewed evidence explicitly says no payment card is required. Re-check the signup flow because eligibility and billing controls can change after the snapshot. The equivalent paid value shown here prices a documented allowance where possible; it is not cash, guaranteed savings, or protection from overage.
Re-read the prompt policy for your plan
This record classifies the reviewed handling as “Not private”: The free developer terms permit storage and broad product, service, and underlying-model improvement uses. Provider policies can distinguish free, paid, enterprise, opted-in, and feature-specific traffic, so verify the governing terms for the exact account and route that will receive your data.
Follow the evidence, then re-check it
The record links 10 first-party sources covering the offer, catalog, limits, pricing, terms, privacy, or adoption evidence available to the audit. Prefer the newest governing document or live catalog when sources disagree, and submit a correction when a provider changes a material term.
Match the quota to the workload shape
Translate the published allowance into the traffic pattern you actually expect instead of comparing headline totals alone. A daily token pool can look generous while a low requests-per-minute or concurrency ceiling blocks interactive bursts; a high request limit can still fail a long-context job when input, output, or per-request tokens are capped. Separate prompt tokens from generated tokens, include retries and tool calls, and test the largest realistic payload. If the provider documents more than one limit window, the tightest window at your peak load is the practical ceiling. Research, experimental, and community access can also carry eligibility or fair-use constraints that cannot be modeled as a simple number.
Plan fallback without changing the rules
A fallback should preserve more than API syntax. Confirm that the substitute route supports the required modality, context length, streaming behavior, structured output, tools, and safety controls, then compare its prompt-retention and training terms. An OpenAI-compatible request shape does not make providers operationally or contractually equivalent. Decide which errors may trigger a retry, cap retry storms, and prevent an exhausted free route from silently switching to a billable model. If deterministic output matters, record the model revision and sampling settings; rotating aliases and free-model pools can change behavior even when the endpoint remains available.
Monitor the offer as a dependency
Capture the model ID, response model field, rate-limit headers, usage fields, latency, HTTP status, and any provider request identifier for each test. Watch for authorization failures, quota exhaustion, catalog removal, policy revisions, and dashboard balance changes as separate failure modes. Re-check the provider’s live catalog and governing pages on a schedule proportionate to the workload’s importance, and keep the dated sources that supported your decision. Free capacity is especially suitable for prototypes, evaluations, fallbacks, and bounded workloads when the application can tolerate change; a production dependency still needs observability, an exit path, and an owner responsible for re-verification.
Test one complete request before scaling
Start with the smallest permitted request using the exact credential, model ID, endpoint, and account type you intend to deploy. Record the HTTP status, response headers, usage fields, latency, and any dashboard balance change. Then exercise the failure path: an invalid model, an exhausted quota, or a rate-limit response should fail clearly without silently switching to a paid route. If the provider supports streaming or tool calls, validate those features separately because a free model can expose a narrower capability set than its paid counterpart. Keep a budget ceiling outside the application whenever billing is possible, and avoid sending sensitive data until the observed route matches the reviewed data agreement.