Liquid d1

# Fix list: Liquid d1

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/liquid-d1, the October 2026 research run, assessed 8 October 2026. Grade E, 43.5 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 Liquid d1: 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, 21 out of 100, up to 15.8 more on the total

Why it scored 21: Scored as a hosted model API, the surface an outside agent would call, with the open weights noted. No status page is linked from the site, the docs or the launch posts, and status.liquid.ai did not answer our reader (0). With no page there is no incident history to read, and the hosted model has been public since 22 September 2026 per the API's own model list (5 of 30). No request or token rate limits found. The docs give image limits only, 8 images and a 4.5 MB body (0 of 15). Liquid documents no 429 or overload handling. A decision has no side effects, so a retry is safe, and TypeSafe's SDKs, the documented clients, retry 408, 429 and 5xx with backoff per the Jev listing's check of 1 October (6 of 15). No SLA for the API. The pricing page mentions SLAs only for enterprise licence customers (0). The `d1` model carries no beta or preview label, though it followed an experimental release by about a week (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.

## 2. Security & auth, 35 out of 100, up to 11.4 more on the total

Why it scored 35: Read for a model API. A bearer key from console.liquid.ai, tied to an organisation. No scopes, expiry or rotation found, and revocation is unchecked because the console needs a login (17 of 30). The terms of 30 September 2026 allow Liquid to use submitted content and outputs for training and fine-tuning, and retention is 'as reasonably necessary'. No opt-out or zero-retention option found. Running d1-3B locally keeps data on your hardware (5 of 20). The terms say results are estimates and call for human review of significant decisions, and the skill file says an approval verdict doesn't replace authorisation checks. Nothing covers hostile or injected content in the state. Local use loads custom code with `trust_remote_code=True`, and we found no network or shell calls in the four files we read (5 of 15). Responses report input tokens, and the terms say organisation administrators can see usage. No request logs documented (4 of 15). No security.txt, disclosure policy or bug bounty found, and the terms forbid vulnerability testing without written authorisation. The Vanta trust centre is unread (4 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, 27 out of 100, up to 9.1 more on the total

Why it scored 27: No x402, MPP or L402 (0). The launch post states $0.04 per million input tokens with no output tokens, readable without a login, and the docs give the image token formula. The price sits in a blog post and on gateway pages, not on Liquid's pricing page (15 of 20). A text-only `d1:free` model is on the API's public model list and in the docs' examples. Its limits aren't published and we couldn't check whether the console asks for a card (12 of 20). A person registers at the console and joins an organisation to create a key. The weights download without an account, but other open-weight listings with a hosted route get no onboarding credit for that (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. Schema & documentation, 59 out of 100, up to 6.7 more on the total

Why it scored 59: No OpenAPI file found. api.liquid.ai/openapi.json and two docs paths returned 404. The API follows TypeSafe's System One request, `GET /decisions/v1/models` answers without a key, and llama.cpp's server README specifies the same request for local runs (10 of 25). llms.txt, llms-full.txt and a Markdown twin of every docs page (10). The docs say what d1 is for and list the tasks that need a language model instead. No limitations or failure-mode page (15 of 20). Three question kinds behind a `type` field, 2 to 10 score levels, and a table of image limits. The hosted limits on questions, options and state size aren't stated (11 of 15). curl, Python and TypeScript examples with full responses for every question type. One error message is documented, for images sent to `d1:free`, and no status codes (9 of 15). Model IDs are `d1` and `d1:free` with no version, and no API changelog was found. The weights have commit history on Hugging Face (4 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.

## 5. Agent ergonomics, 63 out of 100, up to 6 more on the total

Why it scored 63: Read as an API an agent calls for a decision, as with Jev and Clef. Answers are a few numbers a question and no output tokens are produced. Each question is billed as its own prompt, with the state and every image counted again, and no caching or batch endpoint was found for the hosted API (16 of 25). The caller fixes the output shape, with several questions and up to 8 images a request (17 of 20). No error codes or bodies documented beyond one message (5 of 20). Calls are stateless and safe to retry, and Liquid's agent skill file says to treat HTTP errors as failures and keep a bounded retry policy. No retry guidance of Liquid's own (14 of 20). `instructions` and `criteria` are the only question fields, and TypeSafe's Python and TypeScript SDKs work by changing the base URL. They don't send images, so image calls need raw HTTP (11 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.

## 6. Transparency & trust, 54 out of 100, up to 4 more on the total

Made of editorial 46, provenance 62.

Why it scored 54: The hosted model is closed under clear terms. The weights and model code are public under the LFM Open Licence v1.0, which is Apache-2.0 with a $10 million revenue limit on commercial use, so not an OSI licence. A plain-language guide explains it, and training code and data aren't published (20 of 30). The terms and the privacy policy, both dated 30 September 2026, agree that content may be used for training. Neither gives a retention period, and no DPA was found. The site's FAQ still says Liquid has no hosted API of its own (14 of 30). A deprecations page for open models without dates, and no policy for the hosted API (6 of 20). The privacy policy says services are hosted in the United States and names Google Analytics and PostHog. No subprocessor list found, and the trust centre is unread (6 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):

- Domain age: liquid.ai, registered 2017-12-16 (8 years) (11 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)
- Privacy policy: read, states 8 of the 8 things a reader expects, and has 1 clause that costs points (8 of 10)
- Status page: not found (0 of 10)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 60 out of 100, up to 3.5 more on the total

Why it scored 60: Read for a model. The open weights were released on 7 October 2026 and the hosted d1 with vision on 5 October (30). The deprecations page lists retired open models with replacements and no notice periods, and the terms let Liquid change or discontinue any model without obligation (4 of 12). The hosted model went from an experimental release to `d1` with vision in about two weeks under unversioned IDs, too short a record to show churn (4 of 8). No API changelog. Support is a Discord server, and six of eight Hugging Face discussions on d1-3B were merged within days, with replies on the two open ones (8 of 15). No SDK of its own. TypeSafe's SDKs (Python 0.7.2 of 26 September, npm 0.6.0) are the documented clients (8 of 15). The d1-3B repository has three commits on main and custom model code, with GGUF builds and upstream llama.cpp support (6 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.

## 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 status.liquid.ai exists. Our reader's connection to it failed, and no status page is linked from the site or docs
- unchecked: the trust centre at trust.liquid.ai, which renders only with JavaScript, so certifications and any subprocessor list are unread
- unchecked: the console at console.liquid.ai, which needs a login, so key revocation, usage logs, spending limits and whether a card is needed are unread
- unchecked: Liquid's GitHub organisation (Liquid4All). GitHub's API refused our reader with a rate limit
- unchecked: hybrid.py and lfm2_vl.py in the d1-3B repository. We read api.py, runner.py, prompt.py and modeling_d1.py
- The limits of `d1:free` and the rate limits of `d1`. Neither is published
- The hosted context window. OpenRouter and Vercel list 65,536 tokens, and Liquid's docs give 32,768 for d1-3B and no figure for the hosted model
- Whether the hosted `d1` is d1-3B. Liquid's pages don't say
- Which path TypeSafe's SDKs reach. The curl examples post to /decisions/v1/systemone, and the SDK examples set only the base URL https://api.liquid.ai
- The first release date. The API's model list gives `d1:free` a release date of 22 September 2026, OpenRouter says 1 October, and the launch post of 5 October refers to an experimental release the week before
- Whether Vercel AI Gateway accepts images. Its model list tags d1 with vision, and the launch post says the gateways are text only for now
- The model card links https://www.liquid.ai/blog/open-d1, which returns 404. The post is at https://www.liquid.ai/blog/d1-open, and a Hugging Face discussion of 8 October reports the wrong link
- The site's FAQ says Liquid doesn't offer a hosted API of its own, which the launch post and the terms contradict
- Liquid's benchmark, latency and cost comparisons are its own, and we didn't reproduce them

## Weaknesses

- The terms of 30 September 2026 allow Liquid to train on API content, with no opt-out or zero-retention option found
- No status page, rate limits, SLA or error reference found for the hosted API
- The LFM Open Licence v1.0 ends free commercial use at $10 million in annual revenue, so the weights aren't open source
- No OpenAPI file, API changelog or versioned model IDs. The hosted models are `d1` and `d1:free`
- No SDK of its own. TypeSafe's SDKs are the documented clients and don't send images

## 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.

- POST to https://api.liquid.ai/decisions/v1/systemone with a `liquid_` key as a Bearer header. This is not a chat-completions endpoint
- Use `d1` for images. `d1:free` is text-only and answers that it does not accept images
- Send images as Base64 data URLs in `images`, at most 8, with the whole JSON body under 4.5 MB. Remote image URLs are refused
- Budget tokens per question. Each question is billed as its own prompt, with its text and every image counted again
- Don't send confidential content to the hosted API, since the terms allow training on it. Run d1-3B locally for that data

## 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.