NVIDIA NeMo Retriever Embedding and Reranking NIMs

# Fix list: NVIDIA NeMo Retriever Embedding and Reranking NIMs

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/nvidia-nemo-retriever, the October 2026 research run, assessed 8 October 2026. Grade C, 61 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on NVIDIA NeMo Retriever Embedding and Reranking NIMs: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reliability, 53 out of 100, up to 9.4 more on the total

Why it scored 53: Local-software reading, for the containers the owner runs. Signed images install from `nvcr.io` with a support matrix naming GPUs, drivers and CPUs (20). The runtime is closed source and no public CI or test suite was found (0 of 25). There is no public issue tracker. The release notes list three known issues for the Embedding NIM at 2.3, among them a Docker health check that reports unhealthy when the NIM is ready, and two for the Reranking NIM (10 of 25). Release 2.0 lists renamed, deprecated and removed environment variables, but the names changed again by 2.3 (`NIM_BIND_ADDR` to `NIM_SERVER_BIND_ADDR`, `NIM_PRECISION` to `NIM_ENGINE_PRECISION`) with no note in the 2.2 or 2.3 notes we read, and no release carries a date (8 of 15). Version 2.3, with production branches named (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Security & auth, 55 out of 100, up to 7.9 more on the total

Why it scored 55: The API takes no credential. NVIDIA's security page says the deployer must add authentication, and an NGC key is needed only to pull images (10 of 30). The service is stateless inference with no write or delete operation, a configurable bind address and built-in TLS, and no access control of its own (12 of 20). It returns vectors and scores, not untrusted content (10). Prometheus metrics, OpenTelemetry traces and JSON logs go to the operator's own stack, with no per-caller log because there are no callers' identities (10 of 15). NVIDIA PSIRT runs coordinated disclosure and publishes bulletins, images are signed and scanned on NGC with VEX documents, and no security.txt or bug bounty was found (13 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 3. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: Proprietary software with a paid licence, so scored on that licence and not by the free self-hosted rule. No x402, MPP or L402 (0). The licensing guide lists $4,500 a GPU a year and $1 a GPU-hour on cloud marketplaces without a login (20). The developer programme allows research, development and testing on up to 16 GPUs free, and build.nvidia.com says its trial endpoints need no card (20). A person has to create an NGC account and key, and accept licence terms in a browser for some models (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 4. Agent ergonomics, 73 out of 100, up to 4.4 more on the total

Why it scored 73: Vectors can be shortened with `dimensions` on three models or packed as int8 or binary, and returned as base64 (20 of 25). One call takes up to 8,192 inputs or 512 passages with a `truncate` setting, but `/v1/ranking` has no top-n field (14 of 20). Errors are JSON with a message that names the bad field and the allowed values (16 of 20). Calls are stateless and safe to repeat, and 503 is described as try again, with no backoff guidance (14 of 20). Two required fields and the OpenAI request shape, so OpenAI clients work with a model-name suffix. NVIDIA publishes no SDK of its own for these services, and its guide uses the LangChain package (9 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Maintenance & community, 57 out of 100, up to 3.8 more on the total

Why it scored 57: The newest image we could date is `nemotron-3-embed-1b`, updated on NGC on 5 August 2026, 64 days before the check, with Embedding 2.3 pushed on 3 August (20 of 30). Three dated pushes fall inside 90 days across the two services (Reranking 2.3.0 on 27 July, Embedding 2.3 on 3 August, the 5 August update), read from the NGC registry because the release notes have no dates (15 of 20). Closed service reading for responsiveness. Release notes and a developer forum exist, and we did not read reply times (6 of 15). No NVIDIA SDK. `langchain-nvidia-ai-endpoints` 1.4.3 of 2 July 2026 sits in the langchain-ai repository (8 of 15). Images are multi-architecture, signed and rescanned on 5 October 2026 (8 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## 6. Schema & documentation, 78 out of 100, up to 3.6 more on the total

Why it scored 78: OpenAPI 3.1 files to download for both services. The embedding file says version 2.2.0 and the reranking file 1.11.0 while the docs are at 2.3 (25). docs.nvidia.com has an llms.txt and a NIM index, but neither links these docs and the `.html.md` form answers 404 (2 of 10). The usage pages say when to send `query` or `passage`, which embedding type suits which case and what each model can't do (15 of 20). Enums on `input_type`, `modality`, `embedding_type`, `truncate` and `dimensions`, length limits and no extra properties allowed (14 of 15). Curl examples with responses and a table of nine error messages, though the OpenAPI file gives errors a description only (12 of 15). Versioned release notes per service, without dates (10 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 7. Transparency & trust, 71 out of 100, up to 2.5 more on the total

Made of editorial 59, provenance 83.

Why it scored 71: Closed runtime under published terms, with model licences named per model and weights for `nvidia/nemotron-3-embed-1b` on Hugging Face (18 of 30). Inputs stay on the deployer's hardware. The licence agreement says software may collect configuration, performance and usage data, and the AI product terms say NIMs that collect telemetry are switched with `NIM_TELEMETRY_MODE`, a variable release 2.0 removed with no replacement. The current environment variable page names no telemetry setting, so the statements don't fully agree (15 of 30). The AI Enterprise lifecycle policy gives one month of support for a feature branch and nine for a production branch, and the end-of-life table dates the NV-EmbedQA-E5-v5 retirement to January 2027 (18 of 20). Telemetry is disclosed in the terms with an opt-out that the 2.x docs no longer list (8 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Terms of service: read, states 5 of the 7 things a reader expects, and has 1 clause that costs points (6.3 of 10)
- Privacy policy: read, states 7 of the 8 things a reader expects (9.3 of 10)
- security.txt: not found (0 of 10)

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: whether the 2.x containers send any telemetry to NVIDIA. Release 2.0 removed `NIM_TELEMETRY_MODE`, the current environment variable page names no telemetry setting, and we did not run a container
- unchecked: the image list for `nvcr.io/nim/nvidia/nemotron-3-embed-1b`, which NGC's API refused (403, not a public artifact). The repository record lists tags 2.2, 2.2.0, 2.2.1, 2.2.2, 2 and latest, while the getting-started guide pulls `:2.3`
- Release dates. The release notes give none, so the dates here are image push dates from the NGC registry
- unchecked: reply times on the NVIDIA developer forum for NIM questions
- unchecked: the NVIDIA API Trial Terms of Service (a PDF) and the rate limits of the hosted trial endpoints, which were not graded
- unchecked: SOC 2 or ISO 27001 coverage. No certification page was read, and the product is software the owner runs
- No product-specific privacy document was found. `provenance.privacy` points at NVIDIA's general privacy policy, which the model cards name as applicable
- The lead's docs URL under `/text-embedding/latest/` redirects to `/embedding/latest/`, and the hosted endpoints the lead left unchecked do exist as a trial service

## Weaknesses

- The API has no authentication and no rate limiting. The security page leaves both to a proxy the deployer runs
- Production use needs an NVIDIA AI Enterprise licence, $4,500 a GPU a year or $1 a GPU-hour on cloud marketplaces, plus the GPU
- Release notes carry no dates, and environment variable names changed between 2.0 and 2.3 without a note in the 2.2 or 2.3 notes we read
- The NVIDIA Software Licence Agreement forbids disclosing benchmark results without written permission, apart from a published exception
- Closed-source runtime with no public issue tracker or test suite, and no llms.txt entry or Markdown pages for these docs

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Send `input_type` as `query` or `passage` on every embedding call. Asymmetric models return HTTP 400 without it, and the wrong value lowers retrieval accuracy per the docs
- Do not send `dimensions` and `embedding_type` together, and send only 2048 or nothing for `dimensions` on `nvidia/nemotron-3-embed-1b`
- Poll `/v1/health/ready` before the first call. The Docker health check can report unhealthy while the NIM is ready, per the 2.3 known issues
- Check the image tag on NGC before pulling. The guide uses `nemotron-3-embed-1b:2.3`, and NGC's record for that image listed tags up to 2.2.2 on 8 October 2026
- Put a proxy with authentication and TLS in front of port 8000, and sort `/v1/ranking` results yourself as the request has no top-n field

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.