Baseten by Baseten

# Fix list: Baseten

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/baseten, the October 2026 research run, assessed 1 October 2026. Grade B, 66.7 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 Baseten: 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, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No machine payment protocol (0). Per-minute GPU prices from $0.01052 (T4) to $0.16633 (B200) published without a login (20). New workspaces get credits and no payment method is needed until they run out, at which point models are deactivated (20). Signup is 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.

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

Why it scored 55: No field selection, limits or summaries found on list endpoints; responses are small resource objects (10). No pagination parameters on list endpoints such as /v1/models or /v1/secrets (5). The inference error page maps each code to a cause and says which to retry with exponential backoff (20). No idempotency keys; only per-deployment retry endpoints and async requests that can be polled (5). Sensible defaults (`min_replica` 0, `max_replica` 1) and official clients in Python (Truss, CLI 1.0.0) and JavaScript (performance client on npm) (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.

## 3. Reliability, 80 out of 100, up to 4 more on the total

Why it scored 80: Atlassian Statuspage at status.baseten.co with incident history (20). The feed lists 21 incidents from 31 July to 29 September 2026. None took a core API down for an hour; the longest inference one was 82 minutes of intermittent 5xx on 0.15 per cent of requests in one US cluster, and the hour-long management API outages on 3 to 5 August were a scheduled database upgrade. We count that as minor incidents only, at a high rate (20). Management API limits published per endpoint (100 a second, 20 a minute for activate and deactivate) and async at 12,000 a minute (15). 429 returns `retry_after` with an instruction to back off on it, and 529 honours Retry-After (15). SLAs are custom on Enterprise only, none published (0). Dedicated deployments are GA (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. Security & auth, 82 out of 100, up to 3.2 more on the total

Why it scored 82: Team keys can be full access, inference-only or metrics-only and scoped to an environment or a model, organisation keys manage other keys over the API, and every key is revocable; keys don't expire, rotation is manual (30). Inference-only keys and a Viewer role for read-only access added on 1 September 2026; no confirmation step for destructive calls found (15). Returns your own model's output, no third-party content (10). Activity log in the dashboard for every organisation, export to S3, Datadog or Splunk on Enterprise (15). Coordinated disclosure to security@baseten.co on the trust centre, SOC 2 Type II and HIPAA, annual penetration test, no bug bounty and no security.txt; the July 2026 token exposure was disclosed publicly with Baseten's approval by the researcher, not by Baseten (12).

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.

## 5. Transparency & trust, 67 out of 100, up to 2.9 more on the total

Made of editorial 58, provenance 75.

Why it scored 67: Hosted service is closed under clear California terms; Truss, the packaging tool, is MIT (20). The docs say inputs, outputs and weights aren't stored by default, async inputs are kept only while processing, and the privacy policy hands customer content to a published DPA; retention periods are given as 'as long as necessary' and audit-log retention is 'contact support' (22). Model API deprecations carry dates but short notice (announced 13 September for 25 September 2026); no deprecation policy for dedicated deployments found (8). Regions can be listed over the API and EU transfers use SCCs, but no subprocessor list found (8).

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: baseten.co, no registry record we could read (0 of 15)
- security.txt: not found (0 of 10)

## 6. Schema & documentation, 84 out of 100, up to 2.6 more on the total

Why it scored 84: A public OpenAPI 3.0 spec for the management API at api.baseten.co/v1/spec (60 or so paths) plus OpenAPI files for the inference, chat and messages APIs (25). llms.txt with a .md twin for every page (10). Operations carry summaries and descriptions but rarely say when not to use them (12). Request bodies are typed with required fields; enums are sparse (10). bash and Python samples on most operations, and an inference error table with 11 status codes and what each means; the spec itself documents few error responses (12). Versioned /v1 paths and a dated changelog with 10 entries in September 2026 (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. Maintenance & community, 90 out of 100, up to 0.9 more on the total

Why it scored 90: Truss 0.18.32 published on PyPI on 28 September 2026 and a changelog entry on 25 September (30). Truss 0.18.29, 0.18.30 and 0.18.32 in September alone, plus 10 dated changelog entries that month (20). Truss has 9 open issues, the recent ones answered within days, but a 24 August request for a private security contact has one reply and two January 2026 deployment issues sit unanswered (18). Truss, the CLI and a JavaScript client are current (15). Release workflow runs in GitHub Actions and Truss supports Python 3.9 to 3.14, but an open issue reports the npm performance client shipping without its platform dependencies since 0.0.10 (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.

## Deductions

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

- -5: a GitHub personal access token for `basetenbot`, exposed in a public Harbor image since March 2023, gave admin and push access to Baseten's main product repository, the GitOps repository that drives its clusters, its Homebrew tap and per-customer private repositories. Reported on 2026-07-13, revoked on 2026-07-14, no misuse found, published with Baseten's approval in September 2026. Deducted less because the fix was quick and documented (https://www.strix.ai/blog/baseten-harbor-github-pat-takeover)

## 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.

- We couldn't find a subprocessor list; it may sit behind the trust centre's request flow.
- Audit-log retention isn't published.
- The management OpenAPI spec documents few error responses; we didn't test whether live errors follow the inference error table.

## Weaknesses

- H100 at $6.50 and A100 at $4.00 an hour, and start-up and idle replica time are billed
- 21 status-page incidents between 31 July and 29 September 2026, mostly single-cluster 5xx
- A bot token with admin access to the product and GitOps repositories sat exposed from March 2023 until July 2026
- No pagination on management list endpoints and no idempotency keys
- No SLA below Enterprise, no bug bounty and no security.txt

## 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.

- Create a team key with inference-only permission for calling models and keep full-access keys out of the agent
- Sleep for `retry_after` seconds on a 429 from api.baseten.co; the activate and deactivate endpoints allow 20 calls a minute
- Retry 429, 503 and 529 with backoff, but treat 500 as a bug in your model code
- Set `scale_down_delay` below the 900-second default or every burst bills 15 idle minutes
- Send payloads over 256 KiB to `/predict`, not `/async_predict`, unless support has raised the async limit

## What the review panel asked for

- Shorten default idle tail
- Publish an SLA below Enterprise
- Add idempotency keys

## 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.