Gemini 4 Carbon is Google's frontier model that excels at deep work in
complex, real-world tasks across domains including coding, finance, legal,
cybersecurity, and science. Gemini 4 Carbon is a Google Frontier Model
subject to Section 34 of the [Service Specific
Terms](https://cloud.google.com/terms/service-terms) (for details, see [Abuse
monitoring](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/abuse-monitoring#google-frontier-models)).

4 Carbon differs from earlier Gemini models in the
following ways:

- **Unified input and output token limit and long decodes** : 4 Carbon has a unified 1,048,576-token limit that includes both input and output tokens (where output includes thinking, function calls, and the final response) across all supported API surfaces (the [`GenerateContent` API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/reference/models/inference.md.txt), the [Interactions API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/interactions.md.txt), and the [OpenAI compatibility Chat Completions API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/start/openai.md.txt)). A single foreground segment generates up to 65,536 output tokens when streaming (or 32,768 tokens when non-streaming). On both the [Interactions API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/interactions.md.txt) and the [`GenerateContent` API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/reference/models/inference.md.txt) (`v1beta1`), you can extend output up to 1,048,576 tokens (minus your input tokens) using long decodes: the Google Gen AI SDK and Agent Studio handle `continuation_token` automatically across segments, or you can use `background=true` on the Interactions API or pass `continuation_token` in a custom continuation loop (known issue: `continuation_token` is supported only in the `v1beta1` API, which the Google Gen AI SDK uses by default). For details, see the [developer's
  guide to Gemini 4 Carbon](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-4-carbon#long-decodes).
- **Thinking levels** : 4 Carbon supports [thinking
  levels](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/thinking) `LOW`, `MEDIUM` (default on `gemini-4-carbon`), and `HIGH` (default on `gemini-4-carbon-cyber-permissive`). Setting `thinking_budget` or `thinking_level="MINIMAL"` returns an HTTP `400` validation error.
- **Sampling parameters** : Custom values for [sampling
  parameters](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/content-generation-parameters) (`temperature`, `top_p`, and `top_k`) are ignored.
- **Context caching** : 4 Carbon supports [implicit
  caching](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/context-cache/context-cache-overview) only (4,096-token minimum), with durable caching enabled by default. Explicit context caching isn't supported.
- **Office and OpenDocument file inputs** : In addition to PDF (`application/pdf`) and plain text (`text/plain`), 4 Carbon accepts Microsoft Office (`.docx`, `.dotx`, `.doc`, `.pptx`, `.ppt`, `.xlsx`, `.xls`) and OpenDocument (`.odt`, `.odp`, `.ods`) files inline or from Cloud Storage on both the [`GenerateContent` API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/reference/models/inference.md.txt) and the [Interactions API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/interactions.md.txt).

> [!NOTE]
> **Note:** To maintain platform reliability, capacity, and security, Gemini 4 Carbon is available to verified enterprise customers in select regions.


[Try in Agent Studio](https://console.cloud.google.com/agent-platform/studio/chat?model=gemini-4-carbon)

[Deploy example app](https://console.cloud.google.com/agent-platform/studio/chat?suggestedPrompt=How+does+AI+work&deploy=true&model=gemini-4-carbon)
[Developer guide](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-4-carbon.md.txt)
[Pricing](https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing)
Note: "Deploy example app" requires a Google Cloud project with billing and Agent Platform API enabled.

| Model ID | `gemini-4-carbon` ||
| Modalities | Text Input and output Image Input only Audio Input only Video Input only ||
| Token limits | Context window | 1,048,576 |
| Token limits | Maximum output tokens | 1,048,576 |
| Capabilities | - [Thinking](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/thinking.md.txt) Supported - [System instructions](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/prompts/system-instruction-introduction.md.txt) Supported - [Structured output](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/control-generated-output.md.txt) Supported - [Context caching](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/context-cache/context-cache-overview.md.txt) Implicit context caching Supported - [RAG Engine](https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/rag-engine/rag-overview.md.txt) Not supported - [Tuning](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/tuning.md.txt) Not supported - [Agentic video understanding](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/video-understanding.md.txt#agentic-video-processing) Supported ||
| Tools | - [Grounding](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/grounding/overview.md.txt) Google Search, Google Maps, Agent Search, Parallel Web Search, Exa Web Search, Web Grounding for Enterprise Supported - [Code execution](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/tools/code-execution.md.txt) Supported - [Function calling](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/tools/function-calling.md.txt) Supported - [URL context](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/url-context.md.txt) Supported - [Computer use](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/computer-use.md.txt) Supported ||
| APIs | - [GenerateContent API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/reference/models/inference.md.txt) Supported - [Interactions API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/interactions.md.txt) Supported - [Gemini Live API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/live-api.md.txt) Not supported - [Count Tokens API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/get-token-count.md.txt) Supported - [Chat Completions API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/migrate/openai/overview.md.txt) Supported ||
| Consumption options | - [Provisioned Throughput](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/provisioned-throughput.md.txt) Supported - [Batch inference](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/batch-inference.md.txt) Supported - [Pay-as-you-go](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deploy/consumption-options.md.txt) Standard PayGo with Usage Tiers, Priority PayGo Supported - [Fixed quota](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/quotas.md.txt) Not supported ||
| Input size limit | - GenerateContent API: 500 MiB compressed - Interactions API: 256 MiB compressed (synchronous or streaming); 9.75 MiB uncompressed (deferred, when background and store are true) ||
| Interactions API URI limit | Up to 100 distinct Cloud Storage (`gs://`) URIs per request across all modalities on the [Interactions API](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/capabilities/interactions.md.txt), including URIs carried forward from earlier turns using `previous_interaction_id`. `generateContent` supports `gs://` URIs up to the per-modality model item limits (up to 6,500 combined multimodal files when including plain-text files). ||
| Technical specifications | **Image** | - Maximum images per prompt: 3,000 (images and documents combined) - Maximum inline part size (GenerateContent API): 256 MB - Maximum file size per file from Cloud Storage: 2 GiB minus 500 KiB - Maximum file size per file from public HTTP or HTTPS URLs: 15 MiB - Maximum file size per file for direct local uploads in the console: 7 MiB inline (up to 50 MiB staged to Cloud Storage) - Maximum file size per file for Google Drive imports in the console: 7 MB (10 MB total) - Default resolution tokens: 1,120 - Supported MIME types: `image/png`, `image/jpeg`, `image/webp`, `image/heic`, `image/heif` |
| Technical specifications |
| Technical specifications |
| Technical specifications |
| Technical specifications |
|---|---|---|
| **Text** | - Maximum number of files per prompt: 3,000 documents (shared with images; up to 6,500 total files when including plain text) - Maximum number of pages per file: 3,000 - Maximum inline part size (GenerateContent API): 256 MB - Maximum file size per file from Cloud Storage: 2 GiB minus 500 KiB - Maximum file size per file from public HTTP or HTTPS URLs: 15 MiB - Maximum file size per file for direct local uploads in the console: 7 MiB inline (up to 50 MiB staged to Cloud Storage) - Maximum file size per file for Google Drive imports in the console: 7 MB (10 MB total) - Default resolution tokens: 560 - OCR for scanned PDFs: Not used by default - Supported MIME types: `application/pdf`, `text/plain`, `application/msword`, `application/vnd.openxmlformats-officedocument.wordprocessingml.document`, `application/vnd.openxmlformats-officedocument.wordprocessingml.template`, `application/vnd.oasis.opendocument.text`, `application/vnd.ms-powerpoint`, `application/vnd.openxmlformats-officedocument.presentationml.presentation`, `application/vnd.oasis.opendocument.presentation`, `application/vnd.ms-excel`, `application/vnd.openxmlformats-officedocument.spreadsheetml.sheet`, `application/vnd.oasis.opendocument.spreadsheet` |
| **Video** | - Maximum video length (with audio): Approximately 45 minutes - Maximum video length (without audio): Approximately 1 hour - Maximum number of videos per prompt: 10 - Maximum inline part size (GenerateContent API): 256 MB - Maximum file size per file from Cloud Storage: 2 GiB minus 500 KiB - Maximum file size per file from public HTTP or HTTPS URLs: 15 MiB - Default resolution tokens per frame: 70 - Supported MIME types: `video/x-flv`, `video/quicktime`, `video/mpeg`, `video/mpegs`, `video/mpg`, `video/mp4`, `video/webm`, `video/wmv`, `video/3gpp` |
| **Audio** | - Maximum audio length per prompt: Approximately 8.4 hours, or up to 1 million tokens - Maximum number of audio files per prompt: 1,000 - Maximum inline part size (GenerateContent API): 256 MB - Maximum file size per file from Cloud Storage: 2 GiB minus 500 KiB - Maximum file size per file from public HTTP or HTTPS URLs: 15 MiB - Speech understanding for: Audio summarization, transcription, and translation - Supported MIME types: `audio/x-aac`, `audio/flac`, `audio/mp3`, `audio/m4a`, `audio/mpeg`, `audio/mpga`, `audio/mp4`, `audio/ogg`, `audio/pcm`, `audio/wav`, `audio/webm` |
| **Parameter defaults** | - Temperature: 1.0 (fixed; custom values ignored) - topP: 0.95 (fixed; custom values ignored) - topK: 64 (fixed; custom values ignored) - candidateCount: 1 (fixed) |
| **[ML processing](https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/data-residency.md.txt)** | - Multi-region: `us`, `eu` |
| **[Provisioned Throughput](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/provisioned-throughput/supported-models.md.txt)** | - Global: `global` - Multi-region: `us`, `eu` |
| **[Standard PayGo](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/standard-paygo.md.txt)** | - Global: `global` - Multi-region: `us`, `eu` |
| **[Priority PayGo](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/priority-paygo.md.txt)** | - Global: `global` - Multi-region: `us`, `eu` |
| **Batch inference** | - Data residency - CMEK - VPC-SC - AXT |
| **Context caching** | - Data residency - CMEK - VPC-SC - AXT |
| See [Security controls](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/security-controls.md.txt) for more information. ||

## Gemini 4 Carbon for cyber

Gemini 4 Carbon for cyber is a version of
Gemini 4 Carbon for defensive cybersecurity use cases, such as
vulnerability discovery, reverse engineering, exploit verification, and patch
generation. Gemini 4 Carbon for cyber is a Google Frontier Cyber
Model subject to Section 31 of the [Service Specific
Terms](https://cloud.google.com/terms/service-terms) (for details, see [Abuse
monitoring](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/abuse-monitoring#google-frontier-models)).

Gemini 4 Carbon for cyber is generally available on
Gemini Enterprise Agent Platform behind an allowlist. To request access, contact your Google
team or apply through the
[Fairwind Program](https://deepmind.google/fairwind-program/). Access to
earlier Gemini Cyber models (such as
Gemini 4 Argon Cyber Permissive or Gemini 3.8 Flash Cyber)
doesn't carry over; you must apply separately.

The following table shows how Gemini 4 Carbon for cyber differs
from Gemini 4 Carbon:

|   | 4 Carbon | 4 Carbon for cyber |
|---|---|---|
| Model ID | `gemini-4-carbon` | `gemini-4-carbon-cyber-permissive` |
| Access | Verified enterprise customers in select regions | Allowlist only ([Fairwind Program](https://deepmind.google/fairwind-program/)) |
| Default thinking level | `MEDIUM` (supports `LOW`, `MEDIUM`, `HIGH`) | `HIGH` (supports `LOW`, `MEDIUM`, `HIGH`) |
| Computer use | Supported | Not supported |
| Supported regions | `global`, `us`, `eu` | `global`, `us` |