Tinker by Thinking Machines Lab

# Fix list: Tinker

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/tinker, the October 2026 research run, assessed 1 October 2026. Grade D, 51.2 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 Tinker: 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, 35 out of 100, up to 13 more on the total

Why it scored 35: The listing's kind is sdk, but the SDK is a thin client for a hosted training service, so we used the hosted checklist; that's a judgement call. No status page found (0) and so no readable incident history (5). The only published limit we found is the in-flight sample cap, raised from 1,000 to 2,000 in SDK 0.22.4; no rate-limit page (10). The SDK retries sampling with stable request IDs since 0.21.0, and a guide titled 'Avoid Client-Side Timeouts and Retries' covers sampling loops (2026-08-03) (15). No SLA found (0). Sign-up is open and paid, but the cookbook README still says 'after our private beta is over' and we found no GA statement (5).

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

Why it scored 20: No machine payment protocol (0). Per-1M-token prices for prefill, cached prefill, sampling and training published without a login, and as JSON (20). No free tier; the quickstart says to add payment details before training, and the console's credit grant redemption (1 September) isn't a public offer (0). Sign-up 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.

## 3. Security & auth, 55 out of 100, up to 7.9 more on the total

Why it scored 55: API keys from the console or `tinker auth login`, revocable, with key verification added in 0.26.2; no per-key scopes found (20). An org, team and project permissions model was documented on 16 July; we didn't find a read-only role (5). Returns your own model's samples and losses, no third-party content (10). `get_audit_log()` in the SDK, opened to admins on 1 September 2026 (15). A valid security.txt to security-reports@thinkingmachines.ai, expiring 2029-07-13; no bug bounty, certifications or advisories found (5).

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, 53 out of 100, up to 7.6 more on the total

Why it scored 53: Sampling takes token limits and logprob options, so responses can be sized; no field selection (15). RestClient filters training runs by `project_id` since 0.22.6, and the console filters sessions by status (10). No documented error codes found (5). Stable request IDs make SDK retries of sampling safe; nothing on retrying an `optim_step` (15). A LoRA client needs only a base model and a rank, but the SDK is Python only (8).

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. Schema & documentation, 70 out of 100, up to 4.9 more on the total

Why it scored 70: No REST reference or OpenAPI; the contract is the typed Python SDK (15). llms.txt and llms-full.txt since 4 September, plus models.json and serverless.json for machine-readable prices since 31 July (10). Reference pages for ServiceClient, TrainingClient, SamplingClient and RestClient, and an SDK cheatsheet for SFT and RL added on 23 September (12). Typed parameter objects such as `AdamParams`, and checkpoint TTLs bounded from 1 hour to 10 years since 0.29.1 (10). The cookbook's recipes are the examples; no page of error types found (8). Semver 0.x releases with a dated changelog that names removals, such as subprocess-isolated sampling in 0.27.1 (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.

## 6. Transparency & trust, 51 out of 100, up to 4.3 more on the total

Made of editorial 30, provenance 71.

Why it scored 51: The cookbook is Apache-2.0 (5), but we found no terms of service for the hosted service on thinkingmachines.ai or in the docs (0). The privacy notice (2025-08-18) says nothing about training data, weights or whether customer data trains Thinking Machines models, and gives retention as 'as long as reasonably necessary'; checkpoints carry user-set TTLs and can be deleted (8). A model deprecations page lists dated retirements (18 models on 12 June, Kimi-K2.5 on 12 July, Qwen3.6-27B on 2 September 2026) and promises to 'aim to give advance notice' by email (12). Processor categories and US processing are stated, with no named list (5).

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: thinkingmachines.ai, no registry record we could read (0 of 15)
- Status page: not found (0 of 10)

## 7. Maintenance & community, 87 out of 100, up to 1.1 more on the total

Why it scored 87: tinker 0.31.0 on PyPI on 2026-09-30 (30). Ten SDK releases between 10 and 30 September alone (20). The cookbook has 22 open issues and 55 open pull requests, and its README defers outside PRs until the private beta ends; we didn't sample reply times (12). The SDK is current (15). The cookbook runs pytest, pyright and recipe smoke tests in CI and tracks torch 2.10 (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.

- We found no terms of service for Tinker; the hosted service may be governed by terms shown only at sign-up.
- Whether Tinker is declared generally available wasn't established; the cookbook README still refers to a private beta.
- The org, team and project permission guide wasn't read, so whether a read-only role exists is unknown.
- Whether customer training data or checkpoints are used by Thinking Machines isn't stated in the privacy notice we read.

## Weaknesses

- LoRA only; no full-parameter training
- Python SDK only, with no REST reference or OpenAPI
- No terms of service, status page or SLA found
- The privacy notice (August 2025) doesn't cover training data or weights
- Standard-context prices rose on 2026-07-17, and there's 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.

- Set `TINKER_API_KEY` and start from the cookbook recipes rather than the raw primitives
- Read the 'Avoid Client-Side Timeouts and Retries' guide before wrapping sampling calls in your own retries; the SDK already retries sampling with stable request IDs
- Save intermediate checkpoints with a TTL between 1 hour and 10 years; storage bills at $0.10 a GB-month until they expire
- Read models.json for current prices before a run; sampling tokens cost more than training tokens on the open models
- Check the model deprecations page before pinning a base model; 18 were retired on 2026-06-12

## What the review panel asked for

- a GA and stability statement
- a status page
- Publish billing terms

## 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.