Historical model (legacy). The rates below are the last recorded rates, not a recommendation for a new deployment.
Current price verification is unresolved or overdue. These are the last recorded rates; check the official source before budgeting. We do not relabel an old check with today's build date.
Gemini 2.5 Pro (>200k tokens) API Pricing
Gemini 2.5 Pro (>200k tokens) costs $2.50 per 1M input tokens and $15.00 per 1M output tokens. These rates were last verified on .
Published rates and billing units
| Billing dimension | USD list price |
|---|---|
| Input / 1M tokens | $2.50 |
| Cached input / 1M tokens | $0.25 |
| Output / 1M tokens | $15.00 |
Verified 2026-08-12 · Official source · Pricing methodology. Taxes, optional services and account-specific discounts are not included.
Cost examples
These examples use published list prices only. They exclude taxes, enterprise discounts, minimum charges, retries, batch discounts, and provider-specific billing rules.
1M input + 1M output
$17.50
A direct per-million comparison against every other text model.
100K input + 25K output
$0.625
A compact chat, summarization, or analysis workload.
10M input + 2M output
$55.00
A monthly production estimate for heavier RAG or agent traffic.
75% cached input + 250K output
$4.56
Cached input is about 90% cheaper than standard input at list price.
Calculator for Gemini 2.5 Pro (>200k tokens)
Enter your own input, output, and cached-token assumptions. This calculator is preloaded with only Gemini 2.5 Pro (>200k tokens), so the result stays focused on this model.
Input cost
$2.50
Output cost
$15.00
Total
$17.50
Price history
No list-price movement across 202 daily snapshots from 2026-04-23 to 2026-11-11.
When to use Gemini 2.5 Pro (>200k tokens)
Good fit
- reasoning-heavy prompts, planning, analysis, and multi-step tool use
- high-volume routing, classification, extraction, summaries, and fallback traffic
- long-document and repository-scale prompts where context window matters
Be careful with
- blind migration without testing your own prompts, latency, and failure modes