Gemini 3.8 Flash pricing, per 1M tokens, in 2026 and from 2027

Every figure captured 6 September 2026 from the official Google pricing page. Prices change, so treat this as a dated snapshot.

$0.75 per 1M input tokens and $3.75 per 1M output tokens is the paid Standard price for Gemini 3.8 Flash through 31 December 2026, and the same page already lists $1.50 and $7.50 from 1 January 2027. Thinking tokens bill as output. Batch is half of Standard on every line, so $0.375 in and $1.875 out this year. Context cache reads cost $0.075 per 1M plus $0.50 per 1M tokens per hour of storage, and both figures also double in 2027.

Google published the model on ai.google.dev with every price stated twice, one figure "through December 31, 2026" and a second "starting January 1, 2027". Every 2027 figure is exactly 2.0 times its 2026 figure. A workload priced today doubles on the calendar with no change in usage at all. OpenRouter listed google/gemini-3.8-flash with a created timestamp of 1788362056, which is 2 September 2026 at 15:14 UTC. Neither source states a release date.

Standard and Batch prices, with the 2027 step

Gemini 3.8 Flash paid tier, USD per 1M tokens, from ai.google.dev captured 6 September 2026

The page also lists a Flex tier at the Batch prices and a Priority tier at $1.35 in and $6.75 out for 2026 ($2.70 and $13.50 from 2027). Storage gets no Batch discount, it's $0.50 per 1M tokens per hour on every paid tier this year.

The free tier lists input, output and context caching as free of charge on Standard. Grounding with Google Search is not available on the free tier. On the paid tier it includes 5,000 free search requests per month, shared across all Gemini 3.x models, then $14 per 1,000 requests, and Google warns that one request to Gemini "may result in one or more queries to Google Search" with each query billed. The Used to improve our products row reads Yes for the free tier and No for the paid tier. Batch and Flex show no free tier at all.

Three worked cost examples

All three use the prices above, computed in a script. Money is rounded to cents, or to four decimals under one cent.

A chat or classification API at 50,000 requests a day

Take 50,000 requests a day, each with 1,200 input tokens and 300 output tokens, for 30 days. That is 1,800M input tokens and 450M output tokens a month. At the 2026 price the input side is 1,800 x $0.75 = $1,350.00 and the output side is 450 x $3.75 = $1,687.50, so the month costs $3,037.50, or $0.0020 per request. At the 2027 price the same month is 1,800 x $1.50 = $2,700.00 plus 450 x $3.75 doubled to $3,375.00, total $6,075.00, or $0.0041 per request. Output is 56 percent of the bill on 20 percent of the tokens, so reply length moves the number more than prompt length does.

The same workload on Batch

If the 50,000 requests can wait for asynchronous processing, Batch halves every line. Input is 1,800 x $0.375 = $675.00 and output is 450 x $1.875 = $843.75, so the month is $1,518.75 at the 2026 price, or $0.0010 per request. At the 2027 Batch price the month is $3,037.50, the same as 2026 Standard. Moving to Batch cancels the price step for one year, and nothing more.

A RAG pipeline with a 200,000 token cached context

Say a retrieval system holds a 200,000 token corpus in the context cache and reads it 2,000 times a day. Cached reads are 2,000 x 200,000 = 400M tokens a day, billed at $0.075 per 1M, so $30.00 a day. Storage is 0.2M tokens x $0.50 x 24 hours = $2.40 a day. The cached total is $32.40 a day, or $972.00 over 30 days. Sending the same 400M tokens uncached at $0.75 per 1M costs $300.00 a day, or $9,000.00 over 30 days. Caching saves $267.60 a day, an 89.2 percent cut on the context portion.

The question tokens and answers cost the same either way and are left out. At the 2027 prices the cached day is $64.80 against $600.00 uncached, the same 89.2 percent saving. Storage for this corpus is $0.10 an hour and one uncached pass over it costs $0.15, so the cache pays for itself in any hour with a single query. See the cost of a RAG pipeline for the rest of the bill.

How 3.8 Flash compares with 3.5 Flash, Flash-Lite and 3.1 Pro

Gemini paid Standard tier, USD per 1M tokens, ai.google.dev captured 6 September 2026

Against 3.5 Flash, the new model is half the input price and 42 percent of the output price this year. The 2027 step lands 3.8 Flash at $1.50 in, level with 3.5 Flash, and $7.50 out, still below the $9.00 of 3.5 Flash. Budget against the 2027 column for anything that outlives December. Flash-Lite stays the cheap option at $0.30 in and $2.50 out, 2.5x and 1.5x below 3.8 Flash in 2026. The 3.1 Pro Preview costs 2.7x as much on input and 3.2x as much on output, and its prompts over 200k tokens bill at $4.00 in and $18.00 out. Example 1 above on 3.5 Flash would cost $6,750.00 a month, on Flash-Lite $1,665.00, and on 3.1 Pro $9,000.00.

The page lists 3.7 Flash and 3.6 Flash at the same numbers as 3.8 Flash, with the same 2027 step, so price does not separate the 3.x Flash generations. All five models report a context length of 1,048,576 tokens on OpenRouter. The older Gemini 2.0 Flash pricing page and the Gemini Flash versus GPT-4o mini comparison cover the earlier generations.

Related pages