# Google Gemini (/providers/gemini)



Google Gemini is the long-context, multimodal, low-friction option. It is the only first-class
provider that does **both** implicit and explicit caching, it has the **largest context window** in
the set at over 1M tokens, and it is the only one that supports **manual cache clearing**. For any
workload with long repeated prompts and no hard requirement for a specific frontier model, Gemini
implicit caching plus long context often delivers the best cost-per-useful-output of the managed
providers.

It is available on both the [managed-provider](/execution/managed-providers) and
[BYOK](/execution/byok) profiles. Requests that resolve here report
`Agent-Resolved-Provider: gemini`.

## Caching economics [#caching-economics]

Gemini gives you two mechanisms, and you can use either or both:

* **Implicit caching** is automatic for requests whose prefix (1,024+ tokens) matches a recent
  request, up to a 75% discount with zero client-side instrumentation. This is the lowest-friction
  cost optimization across all providers - no breakpoints, no cache IDs.
* **Explicit caching** (the Context Caching API) manually caches a content block - a document, an
  instruction set, a tool list - referenced by cache ID in later requests, billed by cache storage
  duration. Use it when the same large block is referenced across hundreds of requests per day.

| Fact                     | Value                      |
| ------------------------ | -------------------------- |
| Cache type               | Both (implicit + explicit) |
| Minimum cacheable prefix | 1,024 tokens               |
| Cache-read discount      | up to 75%                  |
| Default TTL              | 3,600 seconds (1h)         |
| Extended TTL             | 86,400 seconds (24h)       |
| Cached-token reporting   | Yes                        |
| Manual cache clear       | **Supported**              |

<Note>
  Gemini is the one first-class provider here that reports `manual_cache_clear_supported`. That lets a
  managed-provider purge reach a stronger guarantee for Gemini-cached state than the expiry-bound best
  effort other managed providers are capped at. See [retention and purge](/concepts/retention-and-purge).
</Note>

## Batch, long context, and multimodal [#batch-long-context-and-multimodal]

The **Batch API** delivers a 50% cost reduction at up to 24h turnaround for bulk inference with async
result delivery. The **1M+ token context window** lets a task load a full document and skip a RAG
retrieval step entirely. And Gemini has first-class vision, audio, video, and code input - the broker
routes multimodal workloads here by default.

## When the broker routes here [#when-the-broker-routes-here]

<CardGroup cols="2">
  <Card title="Long repeated prompts" icon="wand-magic-sparkles">
    Workloads with prompts of 1,024+ tokens that do not need explicit cache management - implicit
    caching delivers up to 75% off automatically.
  </Card>

  <Card title="Very long context" icon="scroll">
    Loading a whole document into the 1M+ window to avoid building a retrieval pipeline.
  </Card>

  <Card title="Multimodal input" icon="image">
    Vision, audio, video, and code input route here by default.
  </Card>

  <Card title="Large batch jobs" icon="layer-group">
    Bulk async inference at the 50% Batch discount.
  </Card>
</CardGroup>

## At a glance [#at-a-glance]

| Capability           | Value            |
| -------------------- | ---------------- |
| Context window       | 1,048,576 tokens |
| Multimodal input     | Yes              |
| Live search          | No               |
| Dedicated deployment | No               |
| Service tiers        | standard         |
| Data retention       | standard         |
| Regions              | us, global       |

Manifest revision `cap_2026_06_09`. The [capability manifest](/concepts/capability-manifests) records
that Gemini holds the largest context window and the only manual cache clear in the set, both of which
shape where the broker sends long-context and multimodal work.