DeepSeek API by DeepSeek

# Fix list: DeepSeek API

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/deepseek-api, the October 2026 research run, assessed 1 October 2026. Grade D, 47.1 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 DeepSeek API: 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, 30 out of 100, up to 12.3 more on the total

Why it scored 30: Model reading. One Bearer key per balance, no scoping found (20 of 30). The privacy policy, last updated 10 February 2026, says personal data is used to train models, with an opt-out by email to privacy@deepseek.com (0 of 20). Data kept for as long as needed to provide the services, with no period stated and no zero-retention route (5 of 15). Per-call logs or audit logs not found (5 of 15). No security.txt, and no certification or bug bounty found (0 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.

## 2. Payments & pricing, 20 out of 100, up to 10 more on the total

Why it scored 20: No machine payment protocol (0). Per-token prices published without a login and re-read on 2 October, with cache-hit, cache-miss and output rates at peak and off-peak (20). No free tier found. The pricing page mentions a granted balance spent before topped-up funds, but not how one is granted (0). Browser sign-up and a prepaid top-up (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. Schema & documentation, 49 out of 100, up to 8.3 more on the total

Why it scored 49: No OpenAPI document and no llms.txt found (0 and 0). API reference covers both the OpenAI and Anthropic formats. The chat completion reference, read on 2 October, describes each parameter with its range and default, such as `max_tokens` from 1 to 393,216 with 8K by default without thinking, but says little on when not to use one (14 of 20). Strict tool calls are documented, in beta at the `/beta` base URL, with a JSON Schema subset (object, string, number, integer, boolean, array, enum, anyOf, `$ref`) where every property must be required. JSON output mode carries no schema guarantee (10 of 15). The error codes page lists seven codes, each with a cause and a fix, but no example bodies (10 of 15). Dated updates page that names model changes (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.

## 4. Reliability, 65 out of 100, up to 7 more on the total

Why it scored 65: Status page at status.deepseek.com, run on Flashduty, with per-API components and a readable feed (20). The feed lists 32 incidents from 22 July to 2 October. Four touch only web search. Of the rest, 25 are degraded performance and three are partial unavailability of the API. We opened those three on 2 October. 28 July lasted 7 minutes and 2 September 31 minutes, but 14 September (18:30 to 21:30 UTC) left the V4.1 Flash API and chat partly unavailable for three hours. That's one major outage (10). Concurrency caps published per account, 2,500 on Flash and 500 on V4 Pro, and requests that haven't started inference after 10 minutes are closed (15). The rate-limit and error pages document 429, and advise a brief wait and retry on 500 and 503, with no `Retry-After` header or backoff schedule (10 of 15). No SLA found (0). V4 Pro and V4.1 Flash 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.

## 5. Agent ergonomics, 70 out of 100, up to 4.9 more on the total

Why it scored 70: Model reading of the checklist (tool use, structured output, caching, context, batch, SDKs, errors). Tool calls through both request formats, and the API reference documents `tool_choice` as none, auto, required or a named function (20 of 20). Strict schema-enforced tool calls in beta, JSON output without a schema guarantee (10 of 15). Disk cache on by default with cheaper hits, per the listing (15). 1M-token context and up to 384K output (15). No batch API documented (0). No official SDKs, you use the OpenAI or Anthropic client (0). Seven error codes with a fix for each, 402 for an empty balance (10 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. Maintenance & community, 46 out of 100, up to 4.7 more on the total

Why it scored 46: Model reading. V4.1 Flash released on 10 September (30). No stated notice policy. `deepseek-chat` and `deepseek-reasoner` got three months, announced on 24 April for 24 July (4 of 12). Two shutdown dates in the last 90 days, 24 July and the V4 Flash retirement on 10 September, less 4 because the V4 Pro shutdown set for 14 September was announced on 13 August and withdrawn on 10 September (0 of 8). Dated updates page, and the docs list api-service@deepseek.com and a Discord server for support. We didn't test whether either answers (12 of 15). No official SDK repository to sample (0 of 10). No official SDKs (0 of 15, 0 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.

## 7. Transparency & trust, 68 out of 100, up to 2.8 more on the total

Made of editorial 45, provenance 90.

Why it scored 68: Closed service with clear terms under PRC law. The model weights are published, which doesn't change the hosted service (15). The privacy policy of 10 February 2026 is consistent on training by default and storage in the PRC, but publishes no retention period and no DPA (10 of 30). Dated notices on the updates page, undercut by a shutdown announced and withdrawn within a month and a reroute with no prior entry (10 of 20). Data location disclosed as the PRC, third parties named only by category (10 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):

- 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.

- On 2026-09-10 DeepSeek retired V4 Flash and routed `deepseek-v4-flash` and `deepseek-v4-flash-vision-exp` to V4.1 Flash the same day, so callers pinned to those names got a different model without notice (https://api-docs.deepseek.com/updates)

## 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.

- Whether DeepSeek's support channels answer. The docs list api-service@deepseek.com and a Discord server, and the full FAQ at static.deepseek.com renders only with JavaScript
- How a granted balance is given. The pricing page says it's spent first but not who gets one
- Durations of the 25 degraded-performance incidents weren't opened one by one, so the record is scored on the three partial outages
- The listing's detail 'Data retention: for as long as the account exists' paraphrased a policy that says 'as long as necessary to provide our Services'. Corrected in patch

## Weaknesses

- 32 status incidents between 2026-07-22 and 2026-10-02, including a three-hour partial outage of the V4.1 Flash API on 2026-09-14
- Personal data trains models unless you email to opt out, and data is stored in the PRC with no retention period stated
- V4 Flash names were rerouted to V4.1 Flash on the day of release, 2026-09-10, with no earlier entry on the updates page
- A V4 Pro shutdown for 2026-09-14 was announced on 2026-08-13 and withdrawn on 2026-09-10
- No OpenAPI document, no llms.txt, no security.txt, no batch API, no official SDKs and no free tier

## 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.

- Send only data you'd be happy to publish
- Schedule bulk work off-peak for half price. Peak runs 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, except Chinese public holidays
- Set `max_tokens` yourself. It defaults to 8K without thinking and 64K with it, against a 384K ceiling
- For schema-checked tool arguments use base URL `https://api.deepseek.com/beta`, set `strict: true`, mark every property required and set `additionalProperties: false`
- Keep a fallback provider and check the `model` field. 429 carries no Retry-After, and `deepseek-v4-flash` now answers as V4.1 Flash

## What the review panel asked for

- a minimum notice period
- model names that never move
- Announce price changes ahead
- State failed-call billing

## 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.