Replicate video models

# Fix list: Replicate video models

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/replicate-video, the October 2026 research run, assessed 8 October 2026. Grade B, 62.1 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 Replicate video models: 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. Security & auth, 41 out of 100, up to 10.3 more on the total

Why it scored 41: Bearer tokens with the prefix `r8_`, several per account, each named and able to be disabled. No scopes or expiry (20). No read-only or per-model token and no spend cap found. Prepaid credit stops new work at a zero balance unless auto-reload is on (0). Video output is a media file. Model descriptions and READMEs returned through the MCP server are written by third parties, with no injection guidance found (8). Predictions are listed per account with inputs and logs, API data is deleted after an hour, and no per-token usage view was found (8). Replicate scans public GitHub repositories for leaked tokens and disables them, and webhooks can be verified. security.txt returns 404, and no bug bounty or certification is named on the pages we read (5).

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.

## 2. Payments & pricing, 30 out of 100, up to 8.8 more on the total

Why it scored 30: No x402, MPP or L402 (0). Per-second prices for each official video model are published on its model page without a login, and hardware prices on the pricing page (20). No standing free tier. The try-for-free collection lets a new account run a few models, `minimax/video-01` among them, a limited number of times before buying credit (10 of 20). Signup and token creation are a browser flow (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.

## 3. Reliability, 75 out of 100, up to 5 more on the total

Why it scored 75: Graded as a hosted service, on the predictions API. Replicate is a component of Cloudflare's status page with 30 and 90 day history (20). The component shows 5 days with incidents in 90 days, all in September 2026 and all marked minor by Cloudflare. We read three. Some third-party models couldn't scale out for 15 hours 41 minutes on 14 and 15 September, backend services returned intermittent 500s for 1 hour 54 minutes on 24 September, and hot-swapped Flux models stuck for 17 minutes on 28 September. Under our rule that is minor incidents only (20). One of our own page fetches on 8 October got a 502 from replicate.com and succeeded on retry. Limits of 600 prediction creates and 3,000 other requests a minute are published (15). A 429 body states when the limit resets and the error-code page gives retry advice per code, but no `Retry-After` header or idempotency key is documented (10 of 15). The enterprise page mentions SLAs with no published terms, and the terms of service disclaim service levels (0). The predictions API and official models are generally available (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.

## 4. Agent ergonomics, 72 out of 100, up to 4.6 more on the total

Why it scored 72: A prediction returns a small JSON object with the output as a file URL. The MCP server exposes one tool per HTTP operation, 36 in the OpenAPI document, with an experimental local code mode that reduces them to two (16). Cursor pagination on list endpoints, `webhook_events_filter`, and `Prefer: wait` with a chosen duration (18). A 429 body with the wait time, coded errors with suggested fixes and an `error` field on failed predictions (15). No idempotency key. Failed runs aren't charged, and `Cancel-After`, a cancel endpoint and webhooks limit polling and runaway jobs. MCP annotations weren't checked (8). The four official video models we read require only `prompt`, and there are official Python and JavaScript clients (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, 48 out of 100, up to 4.6 more on the total

Why it scored 48: The brief counts the last published model for a closed service. `alibaba/wan-3` was created on 15 September 2026, 23 days before this check (30). The public changelog's newest entry is 21 April 2026, 170 days ago, which would score 10 on a strict reading, as the Replicate image listing was scored. No dated changelog entries in the last 90 days, and we could date only one model publication in that period (0). The changelog is stale, replicate-python has 70 open issues and was last pushed on 26 August 2025, and support is a contact form (5 of 25). Official Python, JavaScript and MCP packages exist, but the stable releases are `replicate` 1.4.0 on npm from 17 November 2025, 1.0.7 on PyPI from 27 May 2025 and `replicate-mcp` 0.9.0 from 9 June 2025, with 2.0 pre-releases stopping on 18 December 2025 (10). Package health and CI not audited (3).

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. Transparency & trust, 72 out of 100, up to 2.5 more on the total

Made of editorial 56, provenance 87.

Why it scored 72: Closed platform under terms of service dated 1 April 2026 that name Replicate, LLC. The clients are Apache-2.0 and each model keeps its own licence, for example Apache 2.0 for Wan 2.7 (17). The docs say API inputs, outputs, files and logs are deleted after one hour and website predictions are kept until deleted. The privacy policy gives no retention period, and we found no statement on what the outside providers behind official models (Google, Kuaishou, Runway and others) keep (18). No deprecation policy. The terms say Replicate doesn't promise to keep any model and may discontinue parts of the service without notice, while the official-models page promises a stable interface (6). A subprocessor page lists 17 companies, each located in the United States. The model providers aren't on it (15).

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 7 of the 7 things a reader expects, and has 1 clause that costs points (8 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)

## 7. Schema & documentation, 86 out of 100, up to 2.3 more on the total

Why it scored 86: Public OpenAPI document with 26 paths and 36 operations, and each model publishes its own input and output schema (25). llms.txt and Markdown versions of docs pages (10). The collection page says which model suits which job and the model pages describe each input, though few say when not to use a model (15). Inputs are typed with enums and bounds, such as Veo 3.1 `duration` of 4, 6 or 8 and Kling 3.0 `duration` from 3 to 15. Kling's `multi_prompt` is a JSON array passed as a string (13). Examples throughout and 8 coded errors with fixes. The HTTP error body is shown only for 429 (11). The API is versioned at `/v1` and official models are promised a stable interface, but the public changelog's newest entry is 21 April 2026 (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.

## 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: the September 2026 incidents on the two other days the status component counts. Cloudflare's incident feed reached back only to 21 September, so we read three incidents covering three of the five days.
- unchecked: the MCP server's tool definitions and annotations. The `replicate-mcp` package names no source repository on npm, and we didn't open the package.
- Wan 3.0's billing table showed $0.025, $0.05 and $0.10 a second on 8 October 2026, half the README's $0.05, $0.10 and $0.20, under a banner announcing 30 per cent off for the week. We couldn't tell which price applies after the promotion.
- Whether Cloudflare's certifications and security programme cover Replicate. Nothing on the Replicate pages we read says so.
- What the outside providers behind official models keep of prompts, input images and outputs. Not found in the reviewed documentation.
- Maintenance takes 30 for recency from the newest official model (15 September 2026). The Replicate image listing scored the same line from the changelog date, which would give 10 here and a total of 28.
- Job times per model were not established. Replicate publishes none, and we ran no predictions.

## Weaknesses

- Tokens have no scopes, expiry or spend cap, and any token can run every model and delete account resources
- No SLA is published and the terms disclaim service levels. Replicate may withdraw a model without notice
- No idempotency key on prediction creates, and a 429 carries a wait time in the body with no `Retry-After` header documented
- The public changelog's newest entry is 21 April 2026, and the stable npm and PyPI clients date from November and May 2025
- No security.txt, and no certification is named on the pages we read

## 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.

- Call official models at `POST /v1/models/{owner}/{name}/predictions` with no version, then poll `GET /v1/predictions/{id}` or pass `webhook`. Video jobs usually outlast the 60-second `Prefer: wait` window
- Download the output file within the hour. API predictions lose inputs, outputs and logs after 60 minutes
- Read the model's price tiers before setting `resolution`, `mode` or `generate_audio`. Kling 3.0 runs from $0.168 to $0.42 a second and Veo 3.1 doubles with audio
- Send `Cancel-After` on every create. There is no idempotency key, so a blind retry is a second paid video
- Keep credit above $20 or enable auto-reload. Low balances are throttled, and granted credit with no card is held to 6 requests a minute

## 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.