SpendLint

Gemini API pricing / Gemini 2.5 Flash

Gemini 2.5 Flash pricing

Verified Source Google (Gemini API)USD, paid tier, before tax

Gemini 2.5 Flash costs $0.30 per 1M input tokens and $2.50 per 1M output tokens, with thinking tokens billed as output. Cached input is $0.03 per 1M. Gemini 2.5 Flash-Lite is $0.10 / $0.40. The Batch API bills both at 0.5 times these rates.

Price table

ModelInput / 1MOutput / 1MCache read / 1MBatch inputBatch outputEffective from
Gemini 2.5 Flash$0.30$2.50$0.03 $0.15 $1.25
Gemini 2.5 Flash-Lite$0.10$0.40$0.01 $0.05 $0.20
Gemini 3.5 Flash-Lite$0.30$2.50$0.03 $0.15 $1.25

Gemini 3.5 Flash-Lite is shown because it is the model Google positions as the successor at the same price point. The full Gemini table, including Pro, the long-context tier and promotional rows, is on the provider page.

Monthly cost at three token shapes

1,000 requests per day over 30 days. Interactive rate unless marked Batch. Each row is computed from the price rows above.

Shape (input / output tokens)2.5 Flash2.5 Flash, Batch2.5 Flash-Lite3.5 Flash-Lite
Chat turn (3,000 / 600)$72.00$36.00$16.20$72.00
Document extraction (20,000 / 500)$217.50$108.75$66.00$217.50
Long generation (500 / 2,000)$154.50$77.25$25.50$154.50

Two things follow from the table. First, the output price is 8.3 times the input price, so a generation-heavy workload is dominated by output tokens and any thinking the model does. Second, Flash-Lite is 3.0 times cheaper on input but 6.3 times cheaper on output, so the saving from switching grows with the share of output tokens: it is smallest on extraction and largest on generation. Flash-Lite is never dearer at any shape, so there is no break-even point between them, only a quality decision.

Context caching, worked

A chat turn of 3,000 input and 600 output tokens where 2,500 input tokens are a stable system prefix:

The saving is on input only. Cache storage is billed per hour on top and is not modelled here, so a cache that is written once and read rarely can cost more than it saves.

Questions

Does the Gemini 2.5 Flash output price include thinking tokens?

Yes. Google bills thinking tokens at the output rate, so the $2.50 per 1M output figure covers both the visible answer and the reasoning the model does before it. A prompt that triggers a long thinking phase costs more than the visible answer length suggests, which is the most common reason a Flash bill exceeds the estimate.

How much does the Batch API save on Gemini 2.5 Flash?

Batch requests are billed at 0.5 times the interactive rate for input and output, so the chat-turn example above drops from $72.00 to $36.00 per month. Results arrive asynchronously, usually within 24 hours, so it suits backfills and nightly jobs, not user-facing requests.

When is Gemini 2.5 Flash-Lite the better choice?

Flash-Lite is 3.0 times cheaper on input and 6.3 times cheaper on output, so it saves most on output-heavy work, and it is cheaper at every token shape. The decision is therefore about quality, not cost structure. Use Flash-Lite for classification, extraction and routing where a wrong answer is cheap to detect; use Flash where the answer is shown to a person.

Is Gemini 2.5 Flash still worth using now that 3.5 Flash-Lite exists?

On list price 2.5 Flash costs the same as 3.5 Flash-Lite on every dimension in the current dataset ($0.30 input, $2.50 output, $0.03 cache read), so the choice is a quality and latency question, not a cost one. Check the provider table for the effective dates, because Google has run promotional periods on 3.x models.

How is context caching priced on Gemini 2.5 Flash?

Tokens served from cache are billed at $0.03 per 1M instead of $0.30. In the cache example above, moving 2,500 of 3,000 prompt tokens to a cached prefix cuts the per-request cost from $0.002400 to $0.001725. Google also charges for cache storage per hour, which this page does not model; it matters when the cache is large and idle.

Does the free tier change these numbers?

No. These are paid tier list prices in USD before tax. The free tier has different rate limits and data use terms and is not a production baseline.

Sources and verification

Every figure above was read from the official page or pricing API named here and last verified on . Superseded prices stay in the dataset with an end date; nothing is overwritten. When a price on this page is older than 45 days a warning appears at the top. Raw data: /api/v1/pricing.json.