API model string
gemini-2.5-proContext window
1M+ tokens
Max output not published
- Knowledge cutoff
- not published
- Released
- 2025 (GA)
- Modalities
- text/image/audio/video in, text out
Last verified July 10, 2026
Rate limits by tier
Gemini 2.5 Pro uses project-specific quotas that Google does not publish globally — your active RPM and TPM are shown in AI Studio console. The model is paid-only. Prices were raised on July 2, 2026, and a 10% premium applies to non-global Vertex AI endpoints since July 1, 2026.
| Tier | Requirements | RPM | TPM | RPD | Concurrent | Notes |
|---|---|---|---|---|---|---|
| Free (AI Studio) | No billing required | 0 | 0 | 0 | 0 | Gemini 2.5 Pro is PAID-ONLY since April 1, 2026. There is no free quota. Use Gemini 2.5 Flash if you need a free evaluation tier. |
| PAYG ≤200K tokens | Google Cloud billing account linked to project | not published (project-specific; shown in AI Studio console) | not published | not published | not published | $1.25 in / $10.00 out per MTok. Prices raised July 2, 2026. No surcharge applies when all requests stay under 200K tokens. |
| PAYG >200K tokens | Google Cloud billing account | not published | not published | not published | not published | $2.50 in / $15.00 out per MTok. CRITICAL: the surcharge applies to the ENTIRE prompt once it crosses 200K tokens — not just the tokens above the threshold. Crossing by one token nearly doubles input cost. |
| Batch API | Billing account | async (results within 24h) | not published | not published | not published | 50% off standard pricing: $0.625/$5.00 per MTok (≤200K). Use for non-latency-sensitive workloads. |
| Vertex AI | GCP project + Vertex AI API enabled | not published | not published | not published | not published | Vertex AI pricing differs from AI Studio. 10% non-global endpoint premium effective July 1, 2026. Includes VPC, data residency, enterprise SLA. |
Swipe the table sideways to see every limit column.
- 1.Google does not publish stable RPM or TPM numbers for Gemini 2.5 Pro. Your project's active quotas are shown in AI Studio dashboard.
- 2.The 200K long-context surcharge is the largest single budget trap: it applies to the WHOLE prompt once any token crosses the threshold. Even 200,001 tokens triggers 2× input cost for the entire request.
- 3.Thinking tokens are billed at the output rate. They can multiply output token volume 3–10× for complex reasoning tasks — set a thinking budget or disable when not needed.
- 4.Prices were raised on July 2, 2026. The current rates reflect that increase. Always verify at ai.google.dev/pricing.
Limits verified against the Google docs, July 10, 2026.
Token pricing
What you pay per million tokens (USD). Input and output are billed separately.
Input
$1.25
per 1M tokens
Output
$10.00
per 1M tokens
Cached input
$0.13
per 1M cached
Batch discount
−50%
on batch jobs
- Standard pricing (≤200K token requests): $1.25 input / $10.00 output per MTok. Raised July 2, 2026.
- Long-context pricing (>200K tokens): $2.50 input / $15.00 output per MTok. Applies to the ENTIRE prompt once the threshold is crossed.
- Context cache reads: ~$0.125/MTok (~10% of standard input price). Cache storage billed separately.
- Batch API: 50% off → $0.625 input / $5.00 output per MTok (≤200K). Async, results within 24h.
- Thinking tokens billed at output rate. Grounding with Google Search: first 5,000 prompts/month free, then $14/1K queries.
- Vertex AI non-global endpoints carry a 10% premium effective July 1, 2026.
Side-project / chatbot
$16.25
per month
Assumptions
5M in / 1M out/month, all requests ≤200K tokens
5M × $1.25/1M + 1M × $10.00/1M = $6.25 + $10.00
Document analysis with 70% cache hits
$147.75
per month
Assumptions
60M in / 12M out/month; 42M tokens from cache at $0.125
(18M × $1.25 + 42M × $0.125)/1M + 12M × $10.00/1M = $22.50 + $5.25 + $120.00
Enterprise scale
$1,300
per month
Assumptions
400M in / 80M out/month, average request ≤200K tokens
400M × $1.25/1M + 80M × $10.00/1M = $500 + $800
Run your own numbers
Drag your real monthly token volumes and watch the bill update live — priced against rival models at the same usage.
Estimated gemini-2.5-pro spend
$88/mo
Input: $38
Output: $50
30M in × $1.25 + 5M out × $10.00 = $88
Same volume, priced across models
- Gemini 2.5 FlashCheapest$22
- gemini-2.5-proThis model$88
- Claude 4 Sonnet$110
- GPT-5.5$300
Rivals priced at their published input/output rates for the same monthly volumes. Prompt caching is model-specific, so it is applied to gemini-2.5-pro only. Estimates for comparison; real bills vary with request shape and long-context surcharges.
gemini-2.5-pro vs the alternatives
Gemini 2.5 Pro vs leading multimodal APIs on context window, price, surcharges, and availability.
| Aspect | gemini-2.5-pro | Gemini 2.5 Flash | Claude 4 Sonnet | GPT-5.5 |
|---|---|---|---|---|
| Context window | 1M+ | 1M (Gemini 2.5 Flash) | 1M (Claude 4 Sonnet) | 1M (GPT-5.5) |
| Input $/MTok | $1.25 | $0.30 (Flash) | $2.00 (Sonnet intro) | $5.00 (GPT-5.5) |
| Output $/MTok | $10.00 | $2.50 (Flash) | $10.00 (Sonnet intro) | $30.00 (GPT-5.5) |
| Long-context surcharge | yes (>200K: 2× input, 1.5× output, whole prompt) | none (flat pricing at 1M) | none (flat to 1M) | yes >272K |
| Free tier | none (paid-only since Apr 1, 2026) | yes (~1,500 RPD) | none | none |
| Batch discount | 50% | 50% | 50% | 50% |
| Thinking / reasoning | yes (billed as output) | yes (billed as output) | yes (extended thinking) | yes (o-series) |
| Multimodal (video + audio input) | yes (text/image/audio/video) | yes (text/image/audio/video) | text/image only | limited video |
Swipe the table sideways to see every model.
Hitting a 429? The playbook
The exact errors you'll see
429 Too Many RequestsRESOURCE_EXHAUSTEDQuota exceeded for quota metric 'generate_requests_per_minute'You have exceeded your current quota, please check your plan and billing detailsmodels/gemini-2.5-pro is not found for API version v1betaWhy it happens & how to fix it
RPM quota exceeded
Reduce concurrency in your client. Implement a token-bucket throttle to smooth out bursts. Check your current quota in AI Studio and request an increase via Google Cloud Console → Quotas.
Pro model attempted without billing
Add a Google Cloud billing account to your project. Gemini 2.5 Pro has been paid-only since April 1, 2026. Free-tier attempts return a billing-required error.
Long-context requests consuming more quota units
Prompts over 200K tokens are weighted more heavily against project quota. Keep prompts under 200K where possible — this also prevents the 2× input cost surcharge.
Project-level quota cap hit
Go to Google Cloud Console → IAM & Admin → Quotas → Generative Language API → find the Gemini 2.5 Pro quota → Request increase. Approval typically takes 1–5 business days.
Wrong model string
Ensure you are using 'gemini-2.5-pro', not 'gemini-1.5-pro' (retired) or other retired strings. The model-not-found error looks like a 404, but some clients surface it differently.
Retry strategy
Google AI documentation recommends exponential backoff. Pattern: start at 1 second, double each retry, cap at 64 seconds, add ±20% jitter to avoid thundering-herd. Honor the Retry-After header when present. Use Google's official client libraries (Node: @google/generative-ai; Python: google-generativeai) for built-in retry logic. For sustained high load, request a quota increase rather than relying solely on retry loops.
1// retry.ts — Gemini 2.5 Pro with exponential backoff2// Base URL: https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-pro:generateContent34const API_KEY = process.env.GOOGLE_API_KEY!;5const MODEL = 'gemini-2.5-pro';6const BASE_URL = `https://generativelanguage.googleapis.com/v1beta/models/${MODEL}:generateContent`;78async function generateContent(9 prompt: string,10 systemInstruction?: string,11 maxRetries = 5,12): Promise<unknown> {13 let delay = 1000;14 const body: Record<string, unknown> = {15 contents: [{ role: 'user', parts: [{ text: prompt }] }],16 };17 if (systemInstruction) {18 body.systemInstruction = { parts: [{ text: systemInstruction }] };19 }20 for (let attempt = 0; attempt <= maxRetries; attempt++) {21 const response = await fetch(`${BASE_URL}?key=${API_KEY}`, {22 method: 'POST',23 headers: { 'Content-Type': 'application/json' },24 body: JSON.stringify(body),25 });26 if (response.ok) return response.json();27 if (response.status !== 429 || attempt === maxRetries) {28 const errorText = await response.text();29 throw new Error(`Gemini 2.5 Pro error ${response.status}: ${errorText}`);30 }31 const retryAfter = response.headers.get('Retry-After');32 const jitter = 0.8 + Math.random() * 0.4;33 const waitMs = retryAfter ? parseInt(retryAfter) * 1000 : delay * jitter;34 console.warn(`Rate limited. Retrying in ${Math.round(waitMs)}ms (attempt ${attempt + 1}/${maxRetries})`);35 await new Promise((r) => setTimeout(r, waitMs));36 delay = Math.min(delay * 2, 64_000);37 }38 throw new Error('Max retries exceeded');39}How to raise your limits
The ladder from the starter tier to enterprise — what each rung takes, and what it unlocks.
PAYG (start)
MinutesGoogle Cloud Console → create a billing account → link to your AI Studio project
Unlocks: Pay-per-token access at $1.25/$10.00 per MTok (≤200K)
Higher RPM / quota
1–5 business daysAI Studio → check your current quota → Google Cloud Console → IAM & Admin → Quotas → Generative Language API → Request increase
Unlocks: Custom per-project RPM/TPM ceiling
Vertex AI
Same dayEnable Vertex AI API on your GCP project; use the vertexai Python or Node SDK. Reference: cloud.google.com/vertex-ai/pricing
Unlocks: VPC Service Controls, data residency, enterprise SLA, committed use pricing; note 10% non-global endpoint premium since July 1, 2026
Enterprise
NegotiatedContact Google Cloud Sales at cloud.google.com/contact
Unlocks: Dedicated capacity, custom SLAs, higher quota limits, committed use discounts
Cut your token spend
Context caching for repeated prefixes
90% cost reduction on cached tokens (~$0.125/MTok vs $1.25 standard)Cache system prompts or document prefixes shared across many requests. Cache reads cost ~10% of input price. Essential for RAG pipelines, document Q&A, and multi-turn conversations with a large shared context.
Hard-limit prompt length below 200K tokens
Prevents 2× input cost + 1.5× output surcharge for the ENTIRE requestCount tokens before each API call using the tokenCount endpoint. Truncate or chunk documents to stay under 200K. This is the single largest budget trap in Gemini 2.5 Pro — one extra token can nearly double the cost of a request.
Use Gemini 2.5 Flash for high-volume tasks
75%+ cost reduction ($0.30 vs $1.25 input; $2.50 vs $10.00 output)Flash has the same 1M context window, no long-context surcharge, and a free tier. Route only tasks genuinely requiring Pro-level reasoning to 2.5 Pro. Use Flash as the default; escalate to Pro when quality requires it.
Batch API for non-realtime workloads
50% cost reduction (to $0.625/$5.00 per MTok ≤200K)Use the async Batch API for document processing, bulk extraction, classification, and overnight jobs. Results are returned within 24h.
Control thinking tokens
Thinking can multiply output volume 3–10×, multiplying costs proportionallyDisable thinking mode for tasks that don't require step-by-step reasoning (classification, extraction, simple Q&A). Set explicit thinking budget limits for tasks where some reasoning is needed but full budget is excessive.
Monitor Google Search grounding usage
Grounding costs $14/1K queries above 5,000/month freeSet billing alerts in Google Cloud Console if grounding is enabled. At scale, grounding costs can exceed model inference costs. Measure whether grounding quality improvement justifies the spend for your use case.
Set budget alerts at 50% and 90% of monthly spend
Prevents surprise bills from accidental long-context requests or thinking runawayGoogle Cloud Console → Billing → Budgets & Alerts. Two thresholds let you react before hitting the full budget. Especially important given the 200K surcharge, which can multiply costs unexpectedly.
Frequently asked questions
What are the rate limits for Gemini 2.5 Pro?
Google does not publish stable RPM or TPM numbers for Gemini 2.5 Pro. Your active quota is project-specific and shown in the AI Studio dashboard. To increase limits, go to Google Cloud Console → IAM & Admin → Quotas → Generative Language API and request an increase (typically 1–5 business days).
How much does Gemini 2.5 Pro cost?
As of July 2, 2026 (when prices were raised): $1.25 per million input tokens and $10.00 per million output tokens for requests under 200K tokens. Requests over 200K tokens are priced at $2.50/$15.00 per MTok — and the surcharge applies to the entire prompt. Context cache reads cost ~$0.125/MTok. A 50% batch discount is available for async workloads.
Is Gemini 2.5 Pro free?
No. Gemini 2.5 Pro has been paid-only since April 1, 2026. There is no free tier. If you need a free tier, use Gemini 2.5 Flash, which has approximately 1,500 requests/day free (not guaranteed — shown per project in AI Studio).
What is the 200K token surcharge and how do I avoid it?
When a single request has more than 200K tokens in the prompt, both input and output pricing switch to the long-context rate ($2.50/$15.00 per MTok) for the ENTIRE request — not just the tokens above 200K. To avoid it: count tokens before each call, truncate or chunk documents, and use context caching for repeated prefixes. One token over the threshold can nearly double the cost of a request.
How do I fix a 429 RESOURCE_EXHAUSTED error on Gemini 2.5 Pro?
Add exponential backoff to your retry logic (start 1s, double each attempt, cap 64s, add jitter). Check your active quota in AI Studio. For sustained rate limit issues, request a quota increase in Google Cloud Console → IAM & Admin → Quotas. If you're hitting limits because of free-tier, add billing — 2.5 Pro is paid-only.
Gemini 2.5 Pro vs Gemini 2.5 Flash — which should I use?
Use Flash ($0.30/$2.50 per MTok) for high-volume, cost-sensitive, or latency-sensitive workloads. Flash has the same 1M context window, no long-context surcharge, and a free tier. Use Pro ($1.25/$10.00) only when tasks genuinely require its higher reasoning capability. Routing Pro-quality-required requests to Pro and everything else to Flash is the most cost-effective architecture.
How do I increase Gemini 2.5 Pro quota limits?
Go to Google Cloud Console → IAM & Admin → Quotas → search for Generative Language API → find the Gemini 2.5 Pro quota → click Request increase. Approvals typically take 1–5 business days. For enterprise-grade capacity, contact Google Cloud Sales or use Vertex AI. RapidDev can help scope quota requirements and cost models for production systems — reach out at rapidevelopers.com/contact.
Can Gemini 2.5 Pro process video and audio?
Yes. Gemini 2.5 Pro accepts text, image, audio, and video as input, producing text output. This is a key advantage over Claude 4 Sonnet (text/image only) and most GPT models with limited audio/video support. Input tokens for non-text modalities are charged at the standard per-token rates.
We build AI apps that don't hit rate limits
- Retry, backoff & caching built in
- Multi-provider fallback routing
- Fixed price, you own the code
30-min call. No commitment.