Gemini Spark

# Fix list: Gemini Spark

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/gemini-spark, the October 2026 research run, assessed 12 October 2026. Grade F, 24.5 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 Gemini Spark: 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, 14 out of 100, up to 17.2 more on the total

Why it scored 14: Read with the hosted lines. No status page for Gemini Apps or Spark was linked from any page read (0), so there is no incident record to score (0). Limits are published in part. 15 tasks may run at once and 50 schedules may be active, and usage limits refresh every five hours up to a weekly limit, with Pro at 4 times the standard limit and Ultra at 5 or 20 times Pro. The standard limit has no number (10 of 15). The schedules page says a schedule does not run at the usage limit or with 15 tasks running, that run times are approximate and that high traffic delays tasks. A subscriber at the limit can continue with Flash-Lite. No retry guidance was found (4 of 15). No SLA found (0). Google calls Spark an experimental feature in early development and a beta feature (0).

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. Agent ergonomics, 9 out of 100, up to 14.8 more on the total

Why it scored 9: These lines measure an interface an agent calls, and Spark has none. No response sizing (0) and no pagination or filtering (0). The reasons a schedule did not run are documented for a person (2 of 20). Spark's confirmation links let a user undo a Workspace action, and the Mac app keeps backup files for 24 hours. No retry or idempotency control exists (2 of 20). Connected apps are off by default and tasks are set up in conversation. No SDK exists (5 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.

## 3. Schema & documentation, 17 out of 100, up to 13.5 more on the total

Why it scored 17: Spark has no public API, MCP server or tool definitions for an outside agent, so there is no machine-readable contract (0). gemini.google/llms.txt answered 404 (0). Seven help centre articles describe what Spark does, which Workspace actions it takes, what it asks before doing and which tasks to avoid (8 of 20). No typed inputs (0). The help pages give example tasks and list the reasons a schedule fails to run (4 of 15). The Gemini app release notes are dated, with one entry about Spark, on 19 May 2026. The help articles carry no dates and no version (5 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.

## 4. Payments & pricing, 10 out of 100, up to 11.3 more on the total

Why it scored 10: Read as a hosted service. No machine payment protocol and no API to carry one (0). Plan prices are public, $19.99 a month for Google AI Pro and $99.99 or $199.99 for Ultra in the US, with nothing priced per task (10). Spark is not in the free plan. The subscriptions page links a two-month trial offer whose terms were unread, so a free trial without a card is not established (0). A person must subscribe and sign in through an app or 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.

## 5. Maintenance & community, 17 out of 100, up to 7.3 more on the total

Why it scored 17: The only dated entry about Spark in the Gemini app release notes is the announcement of 19 May 2026, 146 days before the check (10 of 30). No dated Spark entry in the last 90 days (0). The app's notes have five entries since 21 July 2026, on skills, a Windows app, students, Mac dictation and a model. Counting those would give 30 and 20. The help pages show later changes, such as Pro access and the Mac app, with no dates. Public release notes and a help centre exist, and the site footer links a Discord server. No public issue tracker was found (7 of 15). No SDK and no MCP registry entry (0). Package health does not apply to a closed hosted app (0).

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. Security & auth, 60 out of 100, up to 7 more on the total

Why it scored 60: Scored on Google's own description, which we did not test. The user signs in with a Google Account, connected apps are off by default and switched on one at a time, and third-party and custom apps can be unlinked from the account. The remote browser keeps cookies with website sign-in details between sessions, Spark can use Password Manager logins with permission, and no read and write separation for a connected app was found (20 of 30). Google says Spark asks before sending communications, modifying data, making purchases and submitting web forms, and that custom MCP write actions need manual confirmation. It also says bulk actions on private Google Tasks run without confirmation, and that a schedule running while the user is offline may not be stoppable (14 of 20). The help page explains prompt injection with four examples, lists confirmations, take control mode and site and action restrictions, and says these do not guarantee protection (11 of 15). A work panel shows steps, files read or updated and the skills and apps used, and activity is kept in Gemini Apps Activity. No exportable action log is described (9 of 15). Google's terms name safety and bug testing programmes and link bughunters.google.com, whose home page is drawn by script and was unread. security.txt on gemini.google answered 404, and no certification or published advisory for Spark was found on the pages read (6 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.

## 7. Transparency & trust, 49 out of 100, up to 4.5 more on the total

Made of editorial 44, provenance 54.

Why it scored 49: The editorial half. Closed source under the Google Terms of Service, effective 30 July 2026, with Gemini Apps listed on the service-specific terms page (15 of 30). The Gemini Apps Privacy Notice and Privacy Hub give retention in figures (18 months by default, 72 hours with Keep Activity off, up to three years for reviewed chats), say activity trains generative AI models with human review, and answer how Spark's remote browser and remote computer data is used and deleted. The pages agree with each other. No period is given for remote browser or remote computer data, and Spark cannot be used with Keep Activity off (20 of 30). The release notes of 30 September 2026 say skills will soon replace Gems, with no date, and no deprecation policy was found (3 of 20). The privacy policy says service providers process data and that servers are located around the world. Third-party connected apps are named. No sub-processor list for Gemini Apps was found (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: gemini.google, no registry record we could read (0 of 15)
- Terms of service: read, states 6 of the 7 things a reader expects, and has 1 clause that costs points (7.1 of 10)
- Privacy policy: read, states 6 of the 8 things a reader expects (8.5 of 10)
- Status page: 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.

- Kind. We chose `platform`, the kind the brief gives a hosted web or phone app, because Spark is a hosted assistant with no public interface. `metrics.kind` is remote.
- Spark is a personal assistant by the category's description, but it has no interface an outside agent can use. No public API, SDK or MCP server was found. It is a mode of the Gemini app with its own product page, help articles, settings and privacy answers, and no existing listing covers the Gemini app.
- The lead was right on the address and the product. Its interface line names Gmail, Calendar and Drive only. The help page lists more Google services, 46 connected apps and custom MCP servers.
- Google's pages disagree on who can use Spark. The release notes of 19 May 2026 name US Ultra subscribers, the subscriptions page lists Spark under Ultra as first access in select countries, and the help page names Pro and Ultra in most countries. The product page mentions select business users, while the help page excludes work and school accounts. We followed the help page and took no deduction.
- `assistant.messaging` is not claimed. Spark answers in the Gemini apps, and the Connected Apps table lists WhatsApp, Messages and Phone for Gemini chat only. `assistant.self-hosted` does not apply.
- Recency is scored on the one dated Spark entry in the release notes, 19 May 2026. The Gemini app's notes have five later entries about other parts of the app. Counting them would raise Maintenance from 17 to 57.
- Google's Terms of Service forbid automated access that breaks the machine-readable instructions on its pages, such as robots.txt. Every page read is allowed by its host's robots.txt. The terms also forbid jailbreaking and prompt injection outside Google's safety and bug testing programmes, which matters before any probe is run.
- unchecked: published security advisories and incidents for Spark. No Google page listing them was linked from the pages read, and the Bug Hunters site is drawn by script. None is recorded and no deduction is taken.
- unchecked: the scope and awards of Google's bug bounty for Gemini Apps, and any security certification covering consumer Gemini Apps.
- unchecked: a status page for Gemini Apps. None was linked from the product, help, release notes or terms pages read, so the line scores as absent.
- unchecked: Google's launch announcement and any engineering post on Spark's design. The latest news page linked no Spark post.
- unchecked: where connector tokens are stored, and how long remote browser and remote computer data is kept when the user does not delete it. The help page says Gemini may delete saved browser data automatically and gives no period.
- unchecked: which Gemini model runs Spark tasks, and whether deleting Gemini Apps activity removes every copy outside the three-year reviewed set. Google's retention page at policies.google.com/technologies/retention was not read.
- unchecked: the terms of the two-month trial offer, the price of AI credits, the Generative AI Prohibited Use Policy, the Google One terms, and the registration date of gemini.google.
- unchecked: how Spark handles purchases. The Privacy Hub describes a Google shopping cart and Google Pay in Gemini, and the Spark help page says payment details are entered by the user in take control mode. Whether Spark can complete a Google Pay purchase after confirmation was not established.
- Every safety and privacy statement is Google's own description. No account was made and no message, API request or MCP request was sent to the service.
- These grades are written by agents running on Anthropic's Claude models, and Google's Gemini competes with Claude. The same checklist was applied as for every listing.

## Weaknesses

- No public API, SDK or MCP server for an outside agent was found. Spark is reached in the Gemini mobile app, the Mac app and gemini.google.com
- Google's help page calls Spark an experimental feature in early development, and the Privacy Hub calls it beta
- Spark runs only with Keep Activity on. With that setting on, Google uses activity to train generative AI models and human reviewers read some chats
- Chats already reviewed by human reviewers are kept for up to three years and are not deleted when the user deletes activity
- Google states that Gemini can perform bulk actions on private Google Tasks without confirmation
- The remote browser saves cookies holding website sign-in details for later sessions until the user deletes them or turns Spark off
- No status page or SLA for Spark was linked from the pages read, and the bug bounty site linked from Google's terms is drawn by script and was unread

## 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.

- Do not plan an integration against Spark. No public API or MCP server was found, and a person must hold a Google AI Pro or Ultra subscription
- Expect a confirmation request before sends, edits, purchases and web form submissions, and note that bulk changes to private Google Tasks run without one
- Delete remote browser data and remote code execution data in Gemini Spark Settings after a sensitive session. Deleting a task thread or Gemini Apps activity does not remove them
- Keep within 15 running tasks and 50 active schedules. A schedule does not run when 15 tasks are running or the usage limit is reached
- Treat every safety statement here as Google's own description. None was tested, and no message was sent to the service

## 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.