How is Gemini priced?

Per input and output token, with separate rates per model tier. Flash and Flash-Lite target high-volume, latency-sensitive work; Pro targets harder reasoning tasks.

Does the long context window cost more?

Longer prompts mean more input tokens, so yes — context length is a direct cost driver even when the per-token rate is unchanged.

Is Gemini Flash cheap enough for high-volume work?

It is among the lowest per-token options for production work. Model your own volumes against it and against Claude Haiku or GPT-4o-mini before committing.

Google Gemini Pricing: Pro, Flash, and Flash-Lite Rates

Current per-token rates

Gemini 2.5 Pro: $1.25/$10 (≤200K), $2.50/$15 (>200K). Gemini 2.5 Flash: $0.30/$2.50. Flash-Lite: $0.10/$0.40.

Long-context pricing tier

Pro doubles in price above 200K input tokens. Forecast with the higher rate if average prompts exceed 100K — a single long doc can flip the tier.

Context caching

Storage billed per hour per 1M tokens (~$1/hr at Pro tier). Cached reads at ~25% of standard. Worth it when the same context is reused 5+ times per hour.

Worked cost example

Doc-Q&A: 50K context + 300 output, 20K calls/month. Pro uncached ≈ $1,310. Flash ≈ $315. Flash-Lite ≈ $102.

When each tier fits

Pro: reasoning, multimodal, long-doc analysis. Flash: default for production. Flash-Lite: classification, extraction, and bulk pipelines where Flash overshoots on capability.

Mechanics: the 200K context-tier cliff

Gemini 2.5 Pro charges $1.25/$10 per 1M up to 200K input tokens, then jumps to $2.50/$15 above that threshold — for the entire call, not just the excess tokens. A prompt at 199K tokens and one at 201K tokens can differ in cost by roughly double, so long-document workloads need to model against the tier boundary explicitly rather than assuming a flat rate.

Worked example: a document sitting near the tier boundary

20K calls/month, 190K input tokens, 300 output tokens on Gemini 2.5 Pro under the low tier: input = 20K*190K=3.8B tokens... at this scale it's clearer per-call: 190K*$1.25/1M=$0.2375 input, 300*$10/1M=$0.003 output, ≈$0.24/call, $4,808/month. Pushing average input to 210K flips every call to the higher tier: 210K*$2.50/1M=$0.525, output $0.003, ≈$0.53/call, $10,560/month — more than double for a 10% token increase.

Decision checklist for long-context workloads

— Is your average prompt comfortably under 200K tokens, or does it hover near the boundary? — Would trimming context (summarizing history, truncating docs) keep you under the tier? — Does context caching's ~25% read rate offset repeated large-context calls? — Have you modelled worst-case prompts, not just average ones, against the tier cliff?

What this pricing breakdown does not cover

This post covers per-token API rates and the context-tier structure; it does not cover Vertex AI-specific commitment discounts, regional pricing variation, or multimodal (image/video/audio) token conversion, which follow separate conversion tables outside the plain text rates used here.

Frequently asked questions

How is Gemini priced?: Per input and output token, with separate rates per model tier. Flash and Flash-Lite target high-volume, latency-sensitive work; Pro targets harder reasoning tasks.
Does the long context window cost more?: Longer prompts mean more input tokens, so yes — context length is a direct cost driver even when the per-token rate is unchanged.
Is Gemini Flash cheap enough for high-volume work?: It is among the lowest per-token options for production work. Model your own volumes against it and against Claude Haiku or GPT-4o-mini before committing.