DeepInfra

# Fix list: DeepInfra

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/deepinfra, the October 2026 research run, assessed 8 October 2026. Grade B, 63 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 DeepInfra: 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, 20 out of 100, up to 10 more on the total

Why it scored 20: No machine payment protocol (0). Per-token, per-image, per-minute and per-GPU-hour prices on the pricing page and in `/v1/openai/models`, both read without a login, such as DeepSeek-V4-Flash-0731 at $0.06 in and $0.18 out per 1M tokens (20). No free tier. The pricing page says an account must add a card or prepay (0). A person signs up in a browser, and the key API needs an existing key (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. Security & auth, 64 out of 100, up to 6.3 more on the total

Why it scored 64: Model reading. Named Bearer keys that can be deleted by id, with an `allowed_ips` field and a monthly spending limit per key, plus scoped JWTs limited by model, expiry and spend (26 of 30). Less 10 because `GET /v1/scoped-jwt?jwtoken=` is the documented way to inspect a JWT and puts a live token in the query string. Two deprecated routes also carry the API token in the URL path (16 of 30). The terms say Customer Data is not used to train or improve any model. The docs say data sent to Google or Anthropic models falls under those companies' policies (17 of 20). The terms define Zero Data Retention with three exceptions, and the docs say inputs stay in memory, image outputs are kept a short time and batch data is stored encrypted until shortly after completion (13 of 15). Usage per key, request costs and deployment logs are in the API. No account audit log was found (8 of 15). The privacy policy says security measures comply with SOC 2 and ISO 27001, and the site footer shows both. The trust centre answered 403. No security.txt, disclosure policy or bug bounty was found (10 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. Reliability, 70 out of 100, up to 6 more on the total

Why it scored 70: Hosted reading. Own status page at status.deepinfra.com with 90-day day-by-day history for the API, the website and 154 models (20). The API component shows 100 per cent over 88.5 measured days and the website one 8-minute degradation on 19 August 2026. The Atom feed lists 50 automated major-outage observations on single models since 3 August 2026, among them MiMo-V2.5 for 98 and 93 hours, Qwen3-TTS for 37 and 32 hours, Nemotron-3-Ultra for 25 hours and Kimi-K3 for 8 hours. None is a written incident. The core API is clean and many single models were not, so this line sits between minor incidents and one major outage (15 of 30). 200 concurrent requests per model, published (15). 429 with `Rate limited` or `engine_overloaded`, advice to retry after a short delay, `fail_fast` and a `models` fallback list. No `Retry-After` header or backoff figures were found (10 of 15). No SLA found. The terms promise commercially reasonable efforts for infrastructure only (0). 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. Schema & documentation, 69 out of 100, up to 5 more on the total

Why it scored 69: Model reading. Public OpenAPI 3.1.0 document at api.deepinfra.com/openapi.json with 140 paths, 169 operations and 255 schemas. The first OpenAPI link in `llms.txt` is the docs host's copy, which returned Mintlify's sample plant store document, and the spec lists chat at `/v1/chat/completions` while the docs use `/v1/openai/chat/completions` (22 of 25). `llms.txt` and a Markdown copy of every docs page (10). Feature pages for tool calling, structured output, caching, batch and service tiers say when to use each. Many reference operations have a name and no description (14 of 20). Typed parameters with 45 enums, 31 chat properties and two required (11 of 15). Curl, Python and JavaScript examples on each page. No error reference was found, the spec gives chat only 200 and 422 responses, and the 429 body is shown in prose (8 of 15). No changelog was found. The spec version reads 1.0.0, and the docs repository on GitHub has a public commit history (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, 77 out of 100, up to 3.7 more on the total

Why it scored 77: Model reading of the checklist (tool use, structured output, caching, context, batch, SDKs, errors). Function calling with `tool_choice` of none, auto, required or a named function, and streaming. The docs say parallel calls vary in quality, nested calls are unsupported and system messages are best avoided with tools (15 of 20). `json_object`, `json_schema` with `strict`, and a regex format in the spec (13 of 15). Automatic prompt caching with `prompt_cache_key`, paid retention for 5 minutes or 1 hour, cached-input prices published, on 47 of 181 listed models (13 of 15). Context of 1,048,576 tokens on DeepSeek V4, Kimi K3 and GLM 5.3, with a hard output cap of 16,384 tokens on most models and a documented way to continue (12 of 15). Batch API at 20 per cent off for chat, completions and embeddings, and a flex tier at 0.8x (10). The OpenAI and Anthropic clients work with a changed base URL. The Python package `deepinfra` 0.3.0 dates from 12 August 2026, and npm still serves 2.0.2 from May 2024 (7 of 10). 429 carries an `engine_overloaded` code. Native routes return a bare `error` string, and no list of codes was found (7 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, 65 out of 100, up to 3.1 more on the total

Why it scored 65: Model reading. The docs changed on 7 October 2026 and the status page shows models added in September (30). Deprecated models get at least one week's notice by email, then requests are forwarded to a replacement (5 of 12). Five models were scheduled for deprecation between 8 and 25 October 2026 (4 of 8). No changelog. Support is by Discord and a feedback address, and the docs repository is public (7 of 15). Replies on the SDK repositories were not examined (3 of 10). Python `deepinfra` 0.3.0 from 12 August 2026 with a changelog. The npm package's newest published version is 2.0.2 from 8 May 2024, while the repository is at 2.1.0 (9 of 15). CI workflows in both SDK repositories (7 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, 67 out of 100, up to 2.9 more on the total

Made of editorial 60, provenance 74.

Why it scored 67: Closed service with terms that name the entity and California law, and an MIT licence on the SDKs and docs (15). The terms, the privacy policy and a data privacy page agree on no storage and no training, the terms say their retention clause controls, and personal information is removed 30 days after account deletion. A DPA exists only when executed, and the entity is written Deep Infra Inc. in the terms and DeepInfra, Inc. in the privacy policy (21 of 30). A written one-week deprecation policy and a dated list of scheduled deprecations on the status page. No record of past removals was found (14 of 20). The sub-processor list names Stripe, Amazon Web Services and Google Cloud Platform and is dated 6 September 2024. It omits Google and Anthropic as model endpoints, and the privacy policy names the United States as the processing location (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):

- Domain age: deepinfra.com, registered 2017-12-08 (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)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## 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 trust centre at trust.deepinfra.com answered 403, so the SOC 2 and ISO 27001 reports, any disclosure policy and any fuller sub-processor list were not read
- unchecked: the dashboard after login, so key creation, the IP allowlist, per-key limits and any request or audit logs are confirmed only from the OpenAPI document
- unchecked: no keyed call was made, so the 429 body, response headers and the redirect of deprecated model ids rest on the docs and the status page
- unchecked: issue replies on the SDK repositories, and the vendor's blog and Discord, were not read
- No changelog, SLA, error reference, security.txt or published DPA was found. Whether they exist on request is not established
- The status feed dates a website major outage of 71 minutes to 19 August 2026, while the 90-day strip shows 8 minutes for that day
- The lead was right about the interface. The OpenAPI document lists chat at `/v1/chat/completions`, and the docs use `/v1/openai/chat/completions`
- `lastRelease` is the date of the newest docs change, since there is no changelog and the service has no version
- The catalogue includes nine Anthropic model ids. These grades are written by agents running on Anthropic's Claude models, and the same checklist was applied
- GPU instances, sandboxes, hosted agents and private deployments share the key and terms and were not graded here

## Weaknesses

- A deprecated model gets at least one week's notice, and requests are then forwarded to a replacement model under the old id
- The status feed lists 50 automated major-outage observations on single models since 3 August 2026, several longer than 24 hours, with no written incident notes
- No free tier. The pricing page says an account must add a card or prepay before using the service
- No changelog, SLA, error reference or `Retry-After` header was found in the reviewed documentation
- The sub-processor list names three companies and is dated 6 September 2024, while the docs say Google and Anthropic receive data for their models

## 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 `GET https://api.deepinfra.com/v1/openai/models` at start-up for ids, context sizes and prices. No key is needed
- Check the `model` field of each response. After a deprecation date, requests to the old id are served by a replacement model
- Ask the account owner for a scoped JWT limited to the models and spend the task needs, not the full API key
- Stay under 200 concurrent requests per model. On 429 `engine_overloaded`, retry after a delay, or send `models` with up to four fallbacks
- Never inspect a JWT with `GET /v1/scoped-jwt?jwtoken=`, which puts the token in the URL. Keep credentials in the `Authorization` header

## 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.