Hyperbolic

# Fix list: Hyperbolic

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/hyperbolic, the October 2026 research run, assessed 8 October 2026. Grade D, 48 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 Hyperbolic: 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, 15 out of 100, up to 10.6 more on the total

Why it scored 15: No x402, MPP or L402 on the GPU API. The vendor's archived x402 repository covered chat completions on another host (0). Hourly prices per configuration are public at `GET /v2/on-demand/rental-options` without a key, and the marketplace page states a starting price of $3.19 a GPU-hour. There is no public price table, the docs point to the console, and storage prices need a key (15 of 20). A free-tier account cannot rent anything. Renting needs a $5 deposit through Stripe, and the only free credit found is from event codes (0). Signup, email verification, the deposit and key creation happen in a browser (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, 52 out of 100, up to 9.6 more on the total

Why it scored 52: Graded on the hosted lines for the REST API. Status page at status.hyperbolic.ai on UptimeRobot with 90-day uptime per monitor (20). The page draws its data by script from a path its robots.txt closes, so we couldn't read the incident record (5, unreadable history). Rate limits appear as numbers only on the plan cards in the docs, 60 requests a minute on the free tier and 600 on the pro tier, and the terms let Hyperbolic change limits without notice (10 of 15). No 429 response, `Retry-After` or backoff guidance in the spec or docs, and no idempotency key on rental creation (0). The docs state a 99.5 per cent uptime SLA for On-Demand and Reserved with credits reviewed case by case, while the Terms of Service disclaim uptime and any duty to give credits (7 of 10). The v2 API is not marked beta. Managed Kubernetes is (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. Security & auth, 47 out of 100, up to 9.3 more on the total

Why it scored 47: Plain API keys sent as a Bearer header, shown once, revocable, any number per account, and attributed to their creator inside an Organisation. No scopes or expiry found, and no secret in a URL (20 of 30). No read-only key and no confirmation step. A key can rent and terminate machines and can call `DELETE /v2/users/me`. Organisation members cannot deposit funds or see billing history (4 of 20). Responses are infrastructure metadata (10). `GET /v2/api-keys/list` shows each key's creator and revocation time, and rental and transaction histories exist. No audit log was found for self-serve accounts (4 of 15). The docs claim a completed SOC 2 Type I audit with Type II under way and ask for vulnerability reports by email with a 48-hour acknowledgement. No security.txt and no bug bounty were found, and the trust centre is drawn by script (9 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.

## 4. Agent ergonomics, 55 out of 100, up to 7.3 more on the total

Why it scored 55: No field selection. Lists page at 25 by default and 100 at most, and `GET /v2/on-demand/active-rental-summary` returns counts and hourly cost without the rental bodies (13 of 25). `page` and `pageSize` on active rentals, a cursor on history, and filters for GPU type, region and term type. Storage volumes and SSH keys return whole arrays (15 of 20). Errors carry `code`, `message` and `issues`, with named cases such as INVALID_INPUT_FOR_RENTAL_TYPE and REBOOT_ALREADY_IN_PROGRESS (14 of 20). No idempotency key on create calls. A duplicate SSH key is rejected and terminating an ended rental returns RENTAL_NOT_FOUND (5 of 20). A rental needs only `rentalType` and `gpuCount`, with defaults for region, GPU type and SSH key. No official SDK or CLI for the v2 API was found (8 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, 42 out of 100, up to 5.1 more on the total

Why it scored 42: The newest changelog entry is for the week of 5 October 2026 (30). It is the only dated entry on the changelog and its RSS feed, so the three-entry test is not met (0 of 20). A public changelog with RSS, in-app chat and email support with stated hours of Monday to Friday, 06:00 to 17:00 Pacific Time (9 of 15). No current official SDK for the v2 API. The vendor's MCP repository last changed on 21 May 2025 and targets legacy v1 paths (0 of 15). The API serves a spec generated from the running service, and there is no client package to assess (3 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, 77 out of 100, up to 3.7 more on the total

Why it scored 77: Public OpenAPI 3.1 document with 65 operations and 53 schemas, served by the API without a key and matching the docs copy (25). `llms.txt`, every docs page as Markdown and a 188 KB `llms-full.txt` (10). Of the 34 current operations, 17 have a description that only repeats the summary. The rest state behaviour, limits and named error cases, and every deprecated operation names its replacement (10 of 20). Required fields, 89 enums, numeric bounds and a `oneOf` for new or existing storage volumes, but rental IDs accept an integer or a string and some integers carry only machine limits (12 of 15). Five error schemas with examples and 400, 401, 403 and 500 on every operation. No request body has an example and no 429 is listed (8 of 15). Versioned in the path with a dated changelog and RSS feed, which holds one entry (12 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, 61 out of 100, up to 3.4 more on the total

Made of editorial 45, provenance 76.

Why it scored 61: Closed service under Terms of Service dated 24 March 2025 and California law. The terms describe a marketplace in which the contract for compute is between buyer and supplier, which does not match the docs' description of a GPU cloud with an SLA (13 of 30). The Privacy Policy is effective 3 June 2024, gives no retention periods and names Babylon three times in one clause. The terms say user content is not used to train generative models. The docs promise data wiping on termination, a 30-day grace period for volumes after the balance reaches zero and a DPA on request, and no published DPA was found (12 of 30). No written deprecation policy, but legacy paths carry a `Sunset` header with an end date of 1 March 2027 and each deprecated operation names its replacement (14 of 20). Region codes are in the spec and Stripe is named as payment processor. The suppliers that host the machines are not named, and the trust centre's sub-processor list, if any, was unreadable (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: hyperbolic.ai, registered 2021-10-11 (4 years) (7 of 15)
- Terms of service: read, states 7 of the 7 things a reader expects, and has 2 clauses that cost points (6 of 10)
- Privacy policy: read, states 5 of the 8 things a reader expects (7.8 of 10)
- security.txt: not found (0 of 10)

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 2026-10-05, removing the `roles` list from the `GET /v2/users/me` response and changing timestamps to RFC 3339 on the v2 API. Both are recorded in the changelog entry for the week they shipped, and no advance notice was found. Documented, so 3 points (https://www.hyperbolic.ai/docs/changelog).

## 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.

- The lead was right that the docs list no MCP server. It missed HyperbolicLabs/hyperbolic-mcp on GitHub, which last changed on 21 May 2025 and calls legacy v1 marketplace paths. We did not test whether those paths still answer.
- unchecked: the status page's monitors, uptime figures and incident history. status.hyperbolic.ai draws them by script from `/api/`, which its robots.txt closes.
- unchecked: trust.hyperbolic.ai is drawn by script (Vanta), so the SOC 2 status, policies and any sub-processor list were not read. The SOC 2 claims come from the docs.
- unchecked: app.hyperbolic.ai/gpus, where the docs send readers for current rates. We read prices only from the open API endpoint, which listed one configuration.
- unchecked: storage volume prices, which `GET /v2/on-demand/storage-volume-options` returns only with a key.
- The home page says builders can serve through an OpenAI-compatible API, and https://api.hyperbolic.ai/v1/models answers 401, but the docs no longer document inference and https://www.hyperbolic.ai/inference answered 404 while still listed in the sitemap. Whether serverless inference is still sold is not established.
- The docs list B300 capacity. The `gpuType` enum in the API has `h100`, `h200`, `b200` and `rtx-6000-pro` only.
- The 99.5 per cent uptime SLA in the docs has no published terms or remedy, and the Terms of Service place uptime between buyer and supplier.
- Whether customers were told before `roles` was removed from `GET /v2/users/me` is unknown.
- The category was kept as gpu-compute. Only `compute.gpu` is recorded, because no serverless, endpoint, batch or container API was found. Managed Kubernetes is a console feature in beta.

## Weaknesses

- API keys carry no scopes or expiry in the reviewed documentation, and the same key can call `DELETE /v2/users/me`
- `POST /v2/on-demand/rentals` has no idempotency key, so a retried call can start a second billed rental
- No 429 response, `Retry-After` header or backoff guidance was found in the spec or the docs
- A free-tier account cannot rent GPUs or storage. Renting needs a one-time deposit of at least $5 through Stripe
- The Terms of Service describe a marketplace in which uptime is between buyer and supplier, while the docs state a 99.5 per cent uptime SLA
- No official SDK or CLI for the v2 API was found. The vendor's MCP repository last changed in May 2025 and calls the legacy v1 marketplace paths

## 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.

- Read `GET /v2/on-demand/rental-options` first. It needs no key and lists what can be rented now, with `costPerHourCents` per GPU configuration.
- Send `rentalType` and `gpuCount` to `POST /v2/on-demand/rentals`. Region defaults to `us-central-1` and GPU type to `h100`, so set both from the options list.
- List active rentals before retrying a failed create call. There is no idempotency key and every ready rental bills hourly.
- Save an SSH public key with `POST /v2/ssh-keys` before renting. Without `sshPublicKeyIds` the newest saved key is attached.
- Do not create reserved rentals unless asked. They are paid in full up front and cannot be terminated early.

## 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.