Nebius AI Cloud

# Fix list: Nebius AI Cloud

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/nebius-ai-cloud, the October 2026 research run, assessed 8 October 2026. Grade B, 67.2 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 Nebius AI Cloud: 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. Payments & pricing, 20 out of 100, up to 10 more on the total

Why it scored 20: No machine payment protocol (0). Per-GPU-hour, per-vCPU-hour and per-GiB prices published without a login, billed by the second (20). No free tier or trial found; adding a card charges $25 to the account balance, and promo codes need billing details first (0). Signup is a browser flow through a Google, GitHub or Microsoft account, after which a service account can work unattended (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.

## 2. Reliability, 57 out of 100, up to 8.6 more on the total

Why it scored 57: Graded as a hosted service on the REST API. Statuspage at status.nebius.com with component history (20). The incident feed lists 22 incidents between 14 July and 8 October 2026, 14 marked major, among them about 21.5 hours of partial degradation in us-central1 on 19 August (with a published post-mortem), about 12 hours of network timeouts in eu-north1 on 22 July and a power fault on several dozen nodes for about 14.7 hours on 7 and 8 October (0). Resource quotas are published with numbers per region, but no request rate limits were found in the reviewed documentation (5 of 15). `RESOURCE_EXHAUSTED` maps to HTTP 429, errors can carry a `TooManyRequests` detail and a `retry_type` of `CALL`, `UNIT_OF_WORK` or `NOTHING`, and `X-Idempotency-Key` covers modifying calls; no Retry-After or backoff intervals found (12 of 15). SLA of 99.5 per cent a month per virtual machine with credits of 10 to 30 per cent; Serverless AI is not on the list of services with a service level (10). Compute and Serverless AI are generally available; Audit Logs, Monitoring and Logging are in preview (10).

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.

## 3. Agent ergonomics, 77 out of 100, up to 3.7 more on the total

Why it scored 77: Lists take `pageSize` and `pageToken`, gets take a `view` and there are get-by-name calls; no field selection (15 of 25). Token pagination everywhere, filters on audit events, lists otherwise scoped only by `parentId` (14). Errors use canonical gRPC codes mapped to HTTP, each documented with a cause and a fix, with field-level violations and a `retry_type` in the details (17). `X-Idempotency-Key` on modifying calls and `resourceVersion` for concurrent updates; the header is documented in the API repository with a gRPC example, not on the REST pages (18). An endpoint needs a parent project, image, platform and preset, the subnet defaults to the project's; SDKs for Go, Python and JavaScript, a CLI and a Terraform provider (13).

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.

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

Why it scored 78: OpenAPI 3.0.3 at api.nebius.cloud/openapi.json with 602 operations on 443 paths and 1,051 schemas, plus the protobuf definitions in `nebius/api` (25). llms.txt, llms-full.txt, every docs page as Markdown and a docs MCP server (10). Every operation has a description, most of one line such as "Creates an endpoint"; field descriptions state mutual exclusions and defaults but rarely when not to call (12). 186 enums and `required` lists on resource metadata; platform and preset are plain strings (11). The docs give curl examples and a table of 15 status codes with fixes, but the OpenAPI document has no examples and documents only 200 responses (8). Paths carry `v1` or `v1alpha1` with a stated rule that alpha versions can change, and the CLI release notes are dated and generated from the same API; the spec's own version reads `version not set` and there is no API changelog as such (12).

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.

## 5. Security & auth, 81 out of 100, up to 3.3 more on the total

Why it scored 81: User tokens last 12 hours. Service accounts authenticate with an uploaded RSA public key (optional expiry), sign a five-minute JWT and exchange it for a 12-hour Bearer token; access comes from group roles granted on a tenant, a project or a single resource. No secret travels in a URL (28). `auditor` and `viewer` roles, narrow roles such as `compute.instance-power-operator`, and a safe mode on by default in the beta MCP server that blocks update and delete commands; no confirmation step for destructive API calls (15). Returns infrastructure metadata and the output of the owner's own containers (10). Audit Logs record control-plane and data-plane events in CloudEvents form, readable at `/audit/v2/audit-events` and exportable to Object Storage; the service is in preview (13). security.txt valid until 31 December 2027, a trust centre listing SOC 2 Type II with HIPAA, SOC 3, ISO 27001, 27018 and 22301 and CSA STAR Level 1, and published rules for customer security scans; no bug bounty found (15).

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.

## 6. Transparency & trust, 77 out of 100, up to 2 more on the total

Made of editorial 70, provenance 83.

Why it scored 77: Closed service under a Services Agreement published 15 September 2026 and effective 28 September, with nine earlier versions archived; the API definitions, SDKs and MCP server are MIT (15). Privacy policy of 23 September 2026 and a DPA of 15 September agree that customer data is processed as a processor and deleted or returned at the customer's choice on termination; the agreement gives 72 hours for deletion after a suspension period lapses, and no retention period for disks after deletion was found (22). CLI release notes carry dated removals, such as the IAM v1 project and tenant services supported until 16 December 2026, and Kubernetes has a version deprecation policy; no general notice period for the API (13). The sub-processor list names each Nebius entity and outside processor with its location and transfer mechanism, and the docs name ten regions (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):

- Endpoint on the vendor's domain: api.nebius.cloud is not on nebius.com (0 of 15)
- Terms of service: read, states 7 of the 7 things a reader expects, and has 1 clause that costs points (8 of 10)

## 7. Maintenance & community, 82 out of 100, up to 1.6 more on the total

Why it scored 82: Python SDK 0.6.20 on 7 October 2026 and CLI 0.12.287 on 6 October (30). 41 PyPI releases since 10 July and CLI releases most working days (20). Closed service with dated release notes, a support centre in the console and a published support regulation; we did not read GitHub issue response times (10 of 15). Current official SDKs for Go, Python and JavaScript (15). Protobuf definitions published to GitHub within a day of each CLI release and a CI badge on `nebius/api`; we did not check the CI results (7).

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.

## 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.

- No request rate limits were found in the docs index, the REST pages or the API repository. They may exist in pages we did not read.
- `X-Idempotency-Key` is documented in the API repository with a gRPC example. We did not confirm on a live call that the REST gateway honours it.
- unchecked: GitHub issue response times and CI results for nebius/api and the SDK repositories. We read the clones, not the issue tracker.
- unchecked: the retention period for Audit Logs events and for disks after deletion. Neither was found in the pages read.
- The Services Agreement bars competitive analysis or benchmarking and unauthorised probing, which matters before any Anchor Terminal probe runs.
- The lead describes serverless endpoints. Serverless AI endpoints are single container VMs started and stopped by a call, with no autoscaling found, so `compute.serverless` is left out of the capabilities.
- Nebius Token Factory and Tavily are separate Nebius products with their own docs and terms and are not graded here.

## Weaknesses

- Status page lists 14 incidents marked major between 14 July and 8 October 2026, including about 21 hours of partial degradation in us-central1 on 19 August
- No request rate limits with numbers and no Retry-After guidance found in the reviewed documentation
- Serverless AI endpoints run on one container VM that is started and stopped by hand; no autoscaling or scale to zero found
- No free tier or trial found. Adding a card at signup charges $25 to the balance, and signup is a browser flow through Google, GitHub or Microsoft
- The OpenAPI document has no examples, documents only 200 responses and reports its version as `version not set`

## 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.

- Use a service account with an authorised key, then exchange a five-minute RS256 JWT at `https://auth.eu.nebius.com/oauth2/token/exchange` for a 12-hour Bearer token
- Send `X-Idempotency-Key` with a random UUID on every create, update and delete, since a 504 can follow a call that succeeded
- Poll the returned operation (`/ai/v1/endpoints/operations/{id}`) until `status` is set; concurrent operations on one resource are not supported
- Stop or delete endpoints when idle. A stopped endpoint bills nothing, a stopped Devlab or VM still bills for its disk
- Check region support first. Serverless AI is absent from `eu-south1` and `us-north1`, and each GPU platform exists in one to four regions
- Run the beta `nebius/mcp-server` with safe mode on (the default); `nebius_cli_execute` can run any CLI command when `SAFE_MODE=false`

## 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.