Gemini Live API

# Fix list: Gemini Live API

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/gemini-live, the October 2026 research run, assessed 8 October 2026. Grade C, 57.1 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 Gemini Live 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. Reliability, 41 out of 100, up to 11.8 more on the total

Why it scored 41: Graded as a hosted API on the Gemini Developer API. A status page exists at aistudio.google.com/status, but it is drawn by script and we could read no components or history on 8 October (10 of 20, and 5 of 30 for the incident line with no readable record). Session limits are published, 15 minutes for audio and 2 for audio with video without compression, connections of about 10 minutes, and spend limits of $10, $50 and $200 per 10 minutes on Tiers 1 to 3. Per-model request limits and concurrent sessions are shown only in AI Studio (8 of 15). The server sends a GoAway message with `timeLeft` before closing, session resumption handles last 2 hours, and the troubleshooting guide gives exponential backoff with jitter for 429 and 503 (12 of 15). No SLA found for the Developer API (0). `gemini-3.8-live` is stable and generally available since 15 September, but the endpoint is `v1beta`, ephemeral tokens are in preview and three docs pages still label the Live API a preview (6 of 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.

## 2. Security & auth, 48 out of 100, up to 9.1 more on the total

Why it scored 48: Keys created since 28 May 2026 are authorisation keys bound to a service account and restricted to the Gemini API, with IP and origin restrictions, a leak checklist and rejection of unrestricted standard keys. Ephemeral tokens add expiry and use counts (25 of 30). The WebSocket guide's documented method is the API key in a `key` query parameter, and ephemeral tokens in `access_token`, so we deduct 10 (15 of 30). Ephemeral tokens can be locked to a model and configuration and to one session, which keeps system instructions server side. There is no approval step for tool calls, which the client runs itself (12 of 20). The model hears callers and can ground on Google Search, and the best practices page covers guardrails in system instructions but not prompt injection (5 of 15). Opt-in logs exist for billed projects, and we did not confirm they cover Live sessions. Usage is visible per project (8 of 15). security.txt on google.com valid to 2030 with a vulnerability reward programme. No certification naming the Developer API was found (8 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. Payments & pricing, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 (0 of 40). Per-token prices with per-minute equivalents are published without a login (20). The Live models are free of charge on the free tier, and the billing guide says new accounts start there with billing linked only to move to the paid tier (20). A key needs a Google account and AI Studio, so a person signs up in a 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.

## 4. Schema & documentation, 66 out of 100, up to 5.5 more on the total

Why it scored 66: No AsyncAPI or similar file for the WebSocket messages was found. The keyless Discovery document (revision 20261006) types `auth_tokens.create` and five related schemas, `BidiGenerateContentSetup` among them, but not the client and server messages (8 of 25). llms.txt with a Markdown twin of every page (10). The capabilities guide, model comparison table and best practices page say when to use each model, voice activity mode and tool behaviour (16 of 20). The reference at ai.google.dev/api/live lists every message with field types, enums such as `ActivityHandling`, `TurnCoverage` and the sensitivity levels, and required marks, as tables and not as a schema (12 of 15). Python and JavaScript examples on every guide, but no list of WebSocket close codes or Live error responses was found (8 of 15). Dated changelog and versioned model names, less 3 because the tools page omits the 3.8 models, the overview says 70 languages where the capabilities guide lists 99, and preview banners remain after general availability (12 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, 69 out of 100, up to 5 more on the total

Why it scored 69: Scored as an API. Context can be sized with sliding-window compression and a trigger token count, media resolution is configurable and `UsageMetadata` reports token counts (18 of 25). No lists to page. Output controls are turn coverage, optional transcription of either side and the compression window (12 of 20). Server signals an agent can act on are `interrupted`, `generationComplete`, GoAway and tool call cancellation, but there is no documented error catalogue for the socket (10 of 20). Session resumption makes reconnects safe, resuming does not spend a token use, and non-blocking function calls take `SILENT`, `WHEN_IDLE` or `INTERRUPT` scheduling. No idempotency guidance for tool side effects after an interruption (14 of 20). A session needs only a model and a response modality, voice activity detection is automatic, and there are official Python and JavaScript SDKs (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. Transparency & trust, 72 out of 100, up to 2.5 more on the total

Made of editorial 55, provenance 89.

Why it scored 72: Closed service with clear terms, SDKs under Apache-2.0 (15 of 30). The terms, pricing page and zero data retention page agree that free-tier content improves Google products and paid content does not. The terms say abuse logs are kept for a limited period and the abuse monitoring page gives 55 days. Session state is kept up to 24 hours when resumption is on, and Search grounding data 30 days (20 of 30). A deprecations table gives earliest shutdown dates and replacements, with 17 November 2026 for the two older Live previews, about nine weeks after the replacement shipped, and no stated minimum notice (15 of 20). The terms say data may be stored transiently or cached in any country where Google has facilities. The Data Processing Addendum applies to paid services, and no sub-processor list was read (5 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):

- Terms of service: read, states 4 of the 7 things a reader expects, and has 3 clauses that cost points (1.4 of 10)
- Privacy policy: read, states 7 of the 8 things a reader expects, and has 1 clause that costs points (7.3 of 10)

## 7. Maintenance & community, 83 out of 100, up to 1.5 more on the total

Why it scored 83: `gemini-3.8-live` went generally available on 15 September 2026, 23 days before the check (30). Three dated Live entries in the last 90 days, the robotics streaming model on 30 July, `gemini-3.5-transcribe-live` on 26 August and the 3.8 Live models on 15 September (20). Closed service with a changelog updated several times a month. Support channels were not tested (10 of 15). Current official SDKs, `google-genai` v2.29.0 and `@google/genai` v2.28.0, both tagged 7 October with about weekly tags since August (15). Apache-2.0 SDKs with type-check workflows in the Python repository and Node 20 or later stated. CI results were not read (8 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.

## 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 incident history and components on aistudio.google.com/status, which is drawn by script
- unchecked: per-model request limits and concurrent session limits for the Live models, shown only at aistudio.google.com/rate-limit, a path robots.txt disallows
- unchecked: whether opt-in AI Studio logs record Live API sessions
- unchecked: GitHub star counts, because the GitHub API was rate limited
- unchecked: the Live API on Vertex AI (Gemini Enterprise Agent Platform), its SLA, certifications and zero data retention terms. This listing grades the Gemini Developer API
- unchecked: Google's sub-processor list and the Data Processing Addendum text
- Whether Live responses are produced in 70 or 99 languages. The overview and the capabilities guide disagree
- Whether the 24-hour session state retention on the zero data retention page and the 2-hour handle validity on the session page describe the same store
- Go and Java SDK support for Live was not checked

## Weaknesses

- The WebSocket guide authenticates with the API key as a `key` query parameter, and ephemeral tokens as an `access_token` query parameter.
- The status page at aistudio.google.com/status renders in the browser only, so no incident history could be read, and no SLA was found for the Developer API.
- No machine-readable contract for the WebSocket messages was found. The Discovery document types only the setup and token schemas.
- The capabilities, best practices and API reference pages still say the Live API is in preview, and the docs give both 70 and 99 supported languages.
- No telephony. Phone calls need a partner such as Voximplant, LiveKit or Pipecat, and concurrent session limits are shown only inside AI Studio.

## 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.

- Enable `sessionResumption` and keep the newest handle. Connections end after about 10 minutes, and handles stay valid for 2 hours.
- Set `contextWindowCompression` with a sliding window. Without it audio sessions stop at 15 minutes and audio with video at 2 minutes, and every turn re-bills the whole context.
- On `gemini-3.8-live` function calls are non-blocking by default. Set `behavior: BLOCKING` if the model must wait for the tool response.
- Send 16 kHz 16-bit PCM in 20 to 40 ms chunks and discard buffered playback when `interrupted` is true.
- Keep the API key on a server and send it through the SDK. Give browsers an ephemeral token from `POST /v1beta/auth_tokens`, locked with `liveConnectConstraints`.

## 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.