Skip to main content
Watch: Grok Bot Built My AI Video Arena While I Tested 8 Models

Cloud platforms - 37 entries

Vertex AI quotas and rate limits

Google Cloud's official docs, read on November 6, 2026, set quotas for Vertex AI across Standard PayGo Tier 1, Standard PayGo Tier 2, Standard PayGo Tier 3, Standard PayGo Tier 4 and Priority PayGo. For Gemini Pro and Argon models, Standard PayGo Tier 2 is listed at $250 - $2,000 Customer spend (30 days) and 1,000,000 Traffic TPM (org-level, prioritized throughput threshold), and Standard PayGo Tier 4 at 10,000,000 Traffic TPM (org-level, prioritized throughput threshold).

Last verified from docs.cloud.google.com. What the docs leave out

Tiers
5
Tables
28
Entries
37
Verified
Nov 6, 2026

Limits by model, endpoint and plan

Values are copied exactly as the docs show them, units included, in the docs' own order. A dash means the docs list no value for that row.

All 28 tables on this page
  1. Gemini Pro and Argon models (includes Gemini 4 Argon and Gemini 4 Argon Cyber Permissive)
  2. Gemini Pro and Argon models
  3. Gemini Flash and Flash-Lite models (GA tier-eligible: Gemini 3.8 Flash Cyber, 3.8 Flash, 3.7 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.5 Flash, 3.1 Flash-Lite, 2.5 Flash-Lite, 2.5 Flash)
  4. Gemini Flash and Flash-Lite models
  5. Gemini Pro models (Priority PayGo; e.g. gemini-4-argon, gemini-4-argon-cyber-permissive, gemini-3.1-pro-preview, gemini-2.5-pro) and Gemini Flash and Flash-Lite models (Priority PayGo; e.g. gemini-3.8-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-2.5-flash)
  6. Gemini generateContent / streamGenerateContent multimodal input, per project per region, per base model and resolution
  7. base_model: gemini-embedding (gemini-embedding-001), global quota
  8. base_model: gemini-embedding-2, global quota
  9. claude-opus-5-5 (Anthropic's Claude Opus 5.5 on Google Cloud), claude-sonnet-5-5 (Anthropic's Claude Sonnet 5.5 on Google Cloud) and claude-haiku-5-5 (Anthropic's Claude Haiku 5.5 on Google Cloud)
  10. claude-opus-5 (anthropic-claude-opus lineage) and claude-opus-4-8 (anthropic-claude-opus lineage)
  11. claude-opus-4-7
  12. claude-opus-4-5
  13. claude-sonnet-5 (anthropic-claude-sonnet lineage)
  14. claude-3-7-sonnet
  15. claude-3-5-sonnet-v2
  16. claude-3-5-sonnet
  17. claude-3-5-haiku
  18. claude-3-opus
  19. claude-3-haiku
  20. claude-fable-5 (anthropic-claude-fable lineage)
  21. Mistral Medium 3 and Mistral Small 3.1 (25.03)
  22. Codestral 2
  23. Mistral OCR (25.05)
  24. Agent Runtime (metrics use the reasoning_engine prefix), per project per region
  25. RAG Engine on Gemini Enterprise Agent Platform
  26. Gen AI evaluation service (default judge model Gemini 2.5 Flash)
  27. Batch inference (Gemini and non-Gemini models)
  28. Semantic Governance Policy, per project per region

Gemini Pro and Argon models (includes Gemini 4 Argon and Gemini 4 Argon Cyber Permissive)

Standard PayGo Tier 1

Gemini Pro and Argon models (includes Gemini 4 Argon and Gemini 4 Argon Cyber Permissive) limits
MetricStandard PayGo Tier 1
Customer spend (30 days)$10 - $250
Traffic TPM (org-level, prioritized throughput threshold)500,000
  • Applies to global endpoint; per model within family; no separate RPM limit.

Gemini Pro and Argon models

3 tiers

Gemini Pro and Argon models limits
TierCustomer spend (30 days)
Standard PayGo Tier 2$250 - $2,000
Traffic TPM (org-level, prioritized throughput threshold)
1,000,000
Standard PayGo Tier 3$2,000 - $50,000
Traffic TPM (org-level, prioritized throughput threshold)
2,000,000
Standard PayGo Tier 4> $50,000
Traffic TPM (org-level, prioritized throughput threshold)
10,000,000
  • Applies to global endpoint; per model within family; no separate RPM limit.

Gemini Flash and Flash-Lite models (GA tier-eligible: Gemini 3.8 Flash Cyber, 3.8 Flash, 3.7 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.5 Flash, 3.1 Flash-Lite, 2.5 Flash-Lite, 2.5 Flash)

Standard PayGo Tier 1

Gemini Flash and Flash-Lite models (GA tier-eligible: Gemini 3.8 Flash Cyber, 3.8 Flash, 3.7 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.5 Flash, 3.1 Flash-Lite, 2.5 Flash-Lite, 2.5 Flash) limits
MetricStandard PayGo Tier 1
Customer spend (30 days)$10 - $250
Traffic TPM (org-level, prioritized throughput threshold)2,000,000
RPMThere's no separate requests-per-minute (RPM) limit for each tier.
  • Usage tiers do not apply to Gemini Nano Banana 2.1, Gemini Omni 1.2 Flash, Gemini 3.1 Flash Image, Gemini 3.1 Flash-Lite Image, Gemini 3 Pro Image, Gemini 2.5 Flash Image (Standard PayGo without tiers), nor to preview models.

Gemini Flash and Flash-Lite models

3 tiers

Gemini Flash and Flash-Lite models limits
TierCustomer spend (30 days)
Standard PayGo Tier 2$250 - $2,000
Traffic TPM (org-level, prioritized throughput threshold)
4,000,000
Standard PayGo Tier 3$2,000 - $50,000
Traffic TPM (org-level, prioritized throughput threshold)
10,000,000
Standard PayGo Tier 4> $50,000
Traffic TPM (org-level, prioritized throughput threshold)
50,000,000
  • Standard PayGo Tier 2 and Standard PayGo Tier 4: Applies to global endpoint; per model within family.
  • Standard PayGo Tier 3: Page example: "an organization in Tier 3 has a prioritized throughput threshold of 10,000,000 TPM for Gemini 3.5 Flash."

Gemini Pro models (Priority PayGo; e.g. gemini-4-argon, gemini-4-argon-cyber-permissive, gemini-3.1-pro-preview, gemini-2.5-pro) and Gemini Flash and Flash-Lite models (Priority PayGo; e.g. gemini-3.8-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-2.5-flash)

2 scopes - Priority PayGo

Gemini Pro models (Priority PayGo; e.g. gemini-4-argon, gemini-4-argon-cyber-permissive, gemini-3.1-pro-preview, gemini-2.5-pro) and Gemini Flash and Flash-Lite models (Priority PayGo; e.g. gemini-3.8-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-2.5-flash) limits
Model or scopeBaseline throughput limit (org-level, per model)
Gemini Pro models (Priority PayGo; e.g. gemini-4-argon, gemini-4-argon-cyber-permissive, gemini-3.1-pro-preview, gemini-2.5-pro)10,000,000 tokens per minute
Gemini Flash and Flash-Lite models (Priority PayGo; e.g. gemini-3.8-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-2.5-flash)50,000,000 tokens per minute
  • Gemini Pro models (Priority PayGo; e.g. gemini-4-argon, gemini-4-argon-cyber-permissive, gemini-3.1-pro-preview, gemini-2.5-pro): "These limits are available to your organization immediately. There's no ramp-up period". Requires header X-Vertex-AI-LLM-Shared-Request-Type: priority (add X-Vertex-AI-LLM-Request-Type: shared to use only Priority PayGo). Traffic above the limit is served at priority when capacity exists; otherwise downgraded to Standard PayGo (trafficType ON_DEMAND instead of ON_DEMAND_PRIORITY). Billed at a higher per-token rate than Standard PayGo.
  • Gemini Flash and Flash-Lite models (Priority PayGo; e.g. gemini-3.8-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-2.5-flash): Supported on global, us and eu endpoints in general; some models (such as Gemini 2.5) may not support all three.

Gemini generateContent / streamGenerateContent multimodal input, per project per region, per base model and resolution

Gemini generateContent / streamGenerateContent multimodal input, per project per region, per base model and resolution limits
MetricLimit
Image input per minute per base model and resolution (generate_content_image_input_per_base_model_id_and_resolution)34,000,000
Video input per minute per base model and resolution (generate_content_video_input_per_base_model_id_and_resolution)192,000,000
Audio input per minute per base model and resolution (generate_content_audio_input_per_base_model_id_and_resolution)11,000,000
Document input per minute per base model and resolution (generate_content_document_input_per_base_model_id_and_resolution)1,200,000
  • Same limits apply on the global endpoint via the corresponding ..._global metric. Metric prefix aiplatform.googleapis.com/.

base_model: gemini-embedding (gemini-embedding-001), global quota

base_model: gemini-embedding (gemini-embedding-001), global quota limits
MetricLimit
Input tokens per minute (global_embed_content_input_tokens_per_minute_per_base_model)100,000,000
Requests per minute (global_embed_content_requests_per_minute_per_base_model)100,000
online_prediction_requests (predict API, gemini-embedding-001)30,000
  • Tuned model inference shares the same quota as the base model.

base_model: gemini-embedding-2, global quota

base_model: gemini-embedding-2, global quota limits
MetricLimit
Input tokens per minute (global_embed_content_input_tokens_per_minute_per_base_model)200,000,000
Requests per minute (global_embed_content_requests_per_minute_per_base_model)60,000

claude-opus-5-5 (Anthropic's Claude Opus 5.5 on Google Cloud), claude-sonnet-5-5 (Anthropic's Claude Sonnet 5.5 on Google Cloud) and claude-haiku-5-5 (Anthropic's Claude Haiku 5.5 on Google Cloud)

3 scopes

claude-opus-5-5 (Anthropic's Claude Opus 5.5 on Google Cloud), claude-sonnet-5-5 (Anthropic's Claude Sonnet 5.5 on Google Cloud) and claude-haiku-5-5 (Anthropic's Claude Haiku 5.5 on Google Cloud) limits
Model or scopeQPM (Multi-region)
claude-opus-5-5 (Anthropic's Claude Opus 5.5 on Google Cloud)1,000
Input TPM (Multi-region, uncached and cache write)
10,000,000
Output TPM (Multi-region)
1,000,000
QPM (global endpoint)
2,000
Input TPM (global endpoint, uncached and cache write)
20,000,000
Output TPM (global endpoint)
2,000,000
Context length
1,000,000
claude-sonnet-5-5 (Anthropic's Claude Sonnet 5.5 on Google Cloud)1,250
Input TPM (Multi-region, uncached and cache write)
12,500,000
Output TPM (Multi-region)
1,250,000
QPM (global endpoint)
2,500
Input TPM (global endpoint, uncached and cache write)
25,000,000
Output TPM (global endpoint)
2,500,000
Context length
1,000,000
claude-haiku-5-5 (Anthropic's Claude Haiku 5.5 on Google Cloud)1,500
Input TPM (Multi-region, uncached and cache write)
15,000,000
Output TPM (Multi-region)
1,500,000
QPM (global endpoint)
3,000
Input TPM (global endpoint, uncached and cache write)
30,000,000
Output TPM (global endpoint)
3,000,000
Context length
1,000,000
  • Values from the Claude quotas by model and region table.

claude-opus-5 (anthropic-claude-opus lineage) and claude-opus-4-8 (anthropic-claude-opus lineage)

2 scopes

claude-opus-5 (anthropic-claude-opus lineage) and claude-opus-4-8 (anthropic-claude-opus lineage) limits
Model or scopeQPM (Multi-region)
claude-opus-5 (anthropic-claude-opus lineage)1,000
Input TPM (Multi-region, uncached and cache write)
10,000,000
Output TPM (Multi-region)
1,000,000
QPM (global endpoint)
2,000
Input TPM (global endpoint, uncached and cache write)
20,000,000
Output TPM (global endpoint)
2,000,000
Context length
1,000,000
claude-opus-4-8 (anthropic-claude-opus lineage)1,000
Input TPM (Multi-region, uncached and cache write)
10,000,000
Output TPM (Multi-region)
1,000,000
QPM (global endpoint)
2,000
Input TPM (global endpoint, uncached and cache write)
20,000,000
Output TPM (global endpoint)
2,000,000
Context length
1,000,000
  • claude-opus-5 (anthropic-claude-opus lineage): Page example: calls to Claude Opus 4.8 through the global endpoint count against the anthropic-claude-opus bucket "regardless of the specific Opus version that you call".
  • claude-opus-4-8 (anthropic-claude-opus lineage): Docs example: calling Claude Opus 4.8 through the global endpoint counts against the anthropic-claude-opus bucket regardless of Opus version.

claude-opus-4-7

Model page: Claude Opus 4.7

claude-opus-4-7 limits
MetricLimit
QPM (Multi-region)400
Input TPM (Multi-region, uncached and cache write)4,000,000
Output TPM (Multi-region)400,000
QPM (global endpoint)800
Input TPM (global endpoint, uncached and cache write)8,000,000
Output TPM (global endpoint)800,000
Context length1,000,000
  • Listed in the Quotas by model and region table; the page does not say whether it uses shared lineage or per-model quota.

claude-opus-4-5

Model page: Claude Opus 4.5

claude-opus-4-5 limits
MetricLimit
QPM (us-east5 / europe-west1 / asia-southeast1, each)200
Input TPM (each regional endpoint, uncached and cache write)2,000,000
Output TPM (each regional endpoint)200,000
QPM (global endpoint)400
Input TPM (global endpoint, uncached and cache write)4,000,000
Output TPM (global endpoint)400,000
Context length200,000

claude-sonnet-5 (anthropic-claude-sonnet lineage)

Model page: Claude Sonnet 5

claude-sonnet-5 (anthropic-claude-sonnet lineage) limits
MetricLimit
QPM (Multi-region)1,250
Input TPM (Multi-region, uncached and cache write)12,500,000
Output TPM (Multi-region)1,250,000
QPM (global endpoint)2,500
Input TPM (global endpoint, uncached and cache write)25,000,000
Output TPM (global endpoint)2,500,000
Context length1,000,000
  • Values from the Claude quotas by model and region table.

claude-3-7-sonnet

claude-3-7-sonnet limits
MetricLimit
QPM (us-east5)55
TPM (us-east5, uncached input and output)500,000
QPM (europe-west1)40
TPM (europe-west1, uncached input and output)300,000
QPM (global endpoint)35
TPM (global endpoint, uncached input and output)300,000
Context length200,000

claude-3-5-sonnet-v2

claude-3-5-sonnet-v2 limits
MetricLimit
QPM (us-east5)90
TPM (us-east5, input and output)540,000
QPM (europe-west1)55
TPM (europe-west1, input and output)330,000
QPM (global endpoint)25
TPM (global endpoint, input and output)140,000
Context length200,000

claude-3-5-sonnet

claude-3-5-sonnet limits
MetricLimit
QPM (us-east5)80
TPM (us-east5, input and output)350,000
QPM (europe-west1)130
TPM (europe-west1, input and output)600,000
QPM (asia-southeast1)35
TPM (asia-southeast1, input and output)150,000
Context length200,000

claude-3-5-haiku

claude-3-5-haiku limits
MetricLimit
QPM (us-east5)80
TPM (us-east5, input and output)350,000
QPM (europe-west1)90
TPM (europe-west1, input and output)400,000
Context length200,000

claude-3-opus

claude-3-opus limits
MetricLimit
QPM (us-east5)20
TPM (us-east5, input and output)105,000
Context length200,000

claude-3-haiku

claude-3-haiku limits
MetricLimit
QPM (us-east5)245
TPM (us-east5, input and output)600,000
QPM (europe-west1)75
TPM (europe-west1, input and output)181,000
QPM (asia-southeast1)70
TPM (asia-southeast1, input and output)174,000
Context length200,000

claude-fable-5 (anthropic-claude-fable lineage)

Model page: Claude Fable 5

claude-fable-5 (anthropic-claude-fable lineage) limits
MetricLimit
QPM (Multi-region)1,000
Input TPM (Multi-region, uncached and cache write)10,000,000
Output TPM (Multi-region)1,000,000
QPM (global endpoint)2,000
Input TPM (global endpoint, uncached and cache write)20,000,000
Output TPM (global endpoint)2,000,000
Context length1,000,000
  • Claude Fable 5.1 is also listed in the quotas table, with empty quota cells.

Mistral Medium 3 and Mistral Small 3.1 (25.03)

2 scopes

Mistral Medium 3 and Mistral Small 3.1 (25.03) limits
Model or scopeQPM (us-central1 / europe-west4, each)
Mistral Medium 390
TPM (input and output, each region)
315,000
Context length
128,000
Mistral Small 3.1 (25.03)60
TPM (input and output, each region)
200,000
Context length
128,000
  • Mistral Medium 3: Quota applies per region where the model is available.

Codestral 2

Codestral 2 limits
MetricLimit
QPM (us-central1 / europe-west4, each)1,100
Input TPM (each region)1,100,000
Output TPM (each region)110,000
Context length128,000 tokens

Mistral OCR (25.05)

Mistral OCR (25.05) limits
MetricLimit
QPM (us-central1 / europe-west4, each)30
Pages per request30 (1 page = 1 million input tokens and 1 million output tokens)
Context length30 pages

Agent Runtime (metrics use the reasoning_engine prefix), per project per region

Agent Runtime (metrics use the reasoning_engine prefix), per project per region limits
MetricLimit
Create, delete, or update Agent Runtime resources per minute (reasoning_engine_service_write_requests)10
Create, delete, or update Agent Runtime sessions per minute (session_write_requests)100
Get, list, or retrieve Agent Runtime sessions per minute (session_read_requests)10000
Query or StreamQuery Agent Runtime per minute (reasoning_engine_service_query_requests)90
Append event to Agent Runtime sessions per minute (session_event_append_requests)300
Maximum number of Agent Runtime resources (reasoning_engine_service_entities)100
Create, delete, or update Agent Runtime memory resources per minute (memory_bank_write_requests)100
Get, list, or retrieve from Agent Runtime Memory Bank per minute (memory_bank_read_requests)300
Sandbox environment (Code Execution) execute requests per minute1000
Sandbox environment (Code Execution) entities per region1000
Sandbox environment (Code Execution) write requests per minute500
A2A Agent post requests like sendMessage and cancelTask per minute60
A2A Agent get requests like getTask and getCard per minute600
Concurrent live bidirectional connections using the BidiStreamQuery API per minute10
  • Metric prefix aiplatform.googleapis.com/.

RAG Engine on Gemini Enterprise Agent Platform

RAG Engine on Gemini Enterprise Agent Platform limits
MetricLimit
RAG Engine data management APIs (VertexRagDataService requests per minute per region)60 RPM
RetrievalContexts API (VertexRagService retrieve requests per minute per region)600 RPM
base_model: textembedding-gecko (Online prediction requests per base model per minute per region per base_model)1,500 RPM
Concurrent ImportRagFiles requests (limit)3 RPM
Maximum number of files per ImportRagFiles request (limit)10,000

Gen AI evaluation service (default judge model Gemini 2.5 Flash)

Gen AI evaluation service (default judge model Gemini 2.5 Flash) limits
MetricLimit
Gen AI evaluation service requests per minute1,000 requests per project per region
Concurrent evaluation runs20 concurrent evaluation runs per project per region
Gemini throughputDepends on model and consumption option
Request timeout (system limit)60 seconds
  • Judge model consumption is calculated at the organization level.

Batch inference (Gemini and non-Gemini models)

Batch inference (Gemini and non-Gemini models) limits
MetricLimit
Gemini models concurrent batch jobsThere are no predefined quota limits on batch inference for Gemini models.
aiplatform.googleapis.com/textembedding_gecko_concurrent_batch_prediction_jobs4
  • Gemini batch uses a shared pool dynamically allocated; requests may be queued for capacity. Same across all regions.

Semantic Governance Policy, per project per region

Semantic Governance Policy, per project per region limits
MetricLimit
Get or list policy resources per minute600
Create, update, or delete policy resources per minute60
Get policy engine resources per minute600
Update or deprovision policy engine resources per minute60
Number of policy resources per project per location1,000

How to raise your limits

  1. Gemini Pro and Argon models and Gemini Flash and Flash-Lite models

    Custom Tier: "Contact your sales team for more information".

  2. Gemini Pro and Argon models (includes Gemini 4 Argon and Gemini 4 Argon Cyber Permissive), Gemini Pro and Argon models and 2 more

    Automatic promotion based on rolling 30-day org spend on eligible Agent Platform services (including Gemini predictions, Gemini features such as caching, Agent Platform CPU/GPU/TPU predictions, management fees and Provisioned Throughput SKUs).

  3. Gemini Pro models (Priority PayGo; e.g. gemini-4-argon, gemini-4-argon-cyber-permissive, gemini-3.1-pro-preview, gemini-2.5-pro) and Gemini Flash and Flash-Lite models (Priority PayGo; e.g. gemini-3.8-flash, gemini-3.5-flash, gemini-3.5-flash-lite, gemini-3.1-flash-lite, gemini-2.5-flash)

    "If your workload requires a higher limit, contact your sales team." For dedicated capacity use Provisioned Throughput.

  4. Gemini generateContent / streamGenerateContent multimodal input, per project per region, per base model and resolution

    "To request a higher limit for these quotas, contact your Google Cloud account team; they can't be increased from the Google Cloud console."

  5. base_model: gemini-embedding (gemini-embedding-001), global quota, base_model: gemini-embedding-2, global quota and claude-sonnet-5 (anthropic-claude-sonnet lineage)

    Request quota changes through Cloud Quotas (Quotas and System Limits page in the Google Cloud console).

  6. claude-opus-5-5 (Anthropic's Claude Opus 5.5 on Google Cloud), claude-sonnet-5-5 (Anthropic's Claude Sonnet 5.5 on Google Cloud) and 5 more

    Google Cloud console Quotas and system limits page, filter base_model anthropic-claude-sonnet.

  7. claude-opus-4-7, claude-opus-4-5 and 10 more

    "If you want to increase any of your quotas for Agent Platform, you can use the Google Cloud console to request a quota increase."

  8. claude-opus-4-7, claude-opus-4-5 and 10 more

    Quotas can typically be adjusted via Cloud Quotas (request a quota increase in the Google Cloud console).

  9. RAG Engine on Gemini Enterprise Agent Platform

    Quotas adjustable via Cloud Quotas; the two "limits" rows are system limits that can't be changed.

  10. Gen AI evaluation service (default judge model Gemini 2.5 Flash)

    "you might need to file a quota increase request. See View and manage quotas".

  11. Batch inference (Gemini and non-Gemini models)

    Quotas and System Limits page: filter aiplatform.googleapis.com/textembedding_gecko_concurrent_batch_prediction_jobs, click the three dots, Edit quota, enter value, Submit request.

  12. Semantic Governance Policy, per project per region

    Quotas and System Limits page: filter the property (for example aiplatform.googleapis.com/semantic_governance/policy_write_requests), Edit quota, Submit request.

What the docs leave out

  • Generative media quotas (Gemini Omni, Veo, Lyria): no numeric values published; docs only say to view/edit in the console's Quotas and System Limits page using metrics global_generate_content_requests_per_minute_per_project_per_base_model (Gemini Omni, Lyria) and long_running_online_prediction_requests_per_base_model (Veo).
  • Claude Fable 5.1 on Google Cloud: quota cells (QPM, Input TPM, Output TPM, context length) are blank in the Claude quotas table.
  • The current Standard PayGo usage tier for an organization is shown on the Agent Platform Dashboard in the Google Cloud console (needs the Agent Platform Viewer and Billing Account Viewer roles).
  • Per-project Claude quotas are viewed and increased on the Quotas and System Limits page in the Google Cloud console; the Claude quotas page says maximum quotas might vary by account.
  • Gemini models under Standard PayGo: "There's no separate requests-per-minute (RPM) limit for each tier", and traffic isn't strictly capped at the tier's TPM threshold. Usage tiers don't apply to Gemini Nano Banana 2.1, Gemini Omni 1.2 Flash, Gemini 3.1 Flash Image, Gemini 3.1 Flash-Lite Image, Gemini 3 Pro Image, Gemini 2.5 Flash Image, or preview models.
  • Provisioned Throughput (dedicated, assured capacity) is not covered here.
  • 2026-11-06: removed claude-opus-4-1, claude-opus-4 and claude-sonnet-4; they are no longer listed in the Claude quotas by model and region table.

FAQ

What are the Vertex AI quotas and rate limits?

Google Cloud's official docs, read on November 6, 2026, set quotas for Vertex AI across Standard PayGo Tier 1, Standard PayGo Tier 2, Standard PayGo Tier 3, Standard PayGo Tier 4 and Priority PayGo. For Gemini Pro and Argon models, Standard PayGo Tier 2 is listed at $250 - $2,000 Customer spend (30 days) and 1,000,000 Traffic TPM (org-level, prioritized throughput threshold), and Standard PayGo Tier 4 at 10,000,000 Traffic TPM (org-level, prioritized throughput threshold).

How do I raise my Vertex AI quotas?

Custom Tier: "Contact your sales team for more information". Automatic promotion based on rolling 30-day org spend on eligible Agent Platform services (including Gemini predictions, Gemini features such as caching, Agent Platform CPU/GPU/TPU predictions, management fees and Provisioned Throughput SKUs). "If your workload requires a higher limit, contact your sales team." For dedicated capacity use Provisioned Throughput.

Where do these Vertex AI numbers come from?

Every value is copied verbatim from Google Cloud's official documentation at docs.cloud.google.com, read on November 6, 2026. This page lists all 37 entries in 28 tables, each linked to the page it came from.

Sources

Last verified . Every value is copied verbatim from the official pages below, read on that date.

Newsletter

Know when limits change

One email per week. New model releases, pricing and limit changes, and what they mean for what you build.