Blog · 2026-10-23 · By Dhrumil Barot, Founder

Gemini API Pricing: Every Model, Tier and Promo End Date

Gemini API pricing verified 2026-10-23: every model's input, cache and output rate, when Gemini 4 Argon and Flash promo prices double, Batch/Flex/Priority tiers, long context, grounding fees and cost per task.

TL;DR: Gemini API pricing on 2026-10-23 runs from $0.10 input / $0.40 output per million tokens (gemini-2.5-flash-lite) to $4 / $18 (gemini-3.1-pro-preview above 200K tokens). The flagship gemini-4-argon costs $2 / $10 until 2026-11-30, then doubles to $4 / $20. Flash models double on 2027-01-01. Batch and Flex halve every rate.

Gemini API pricing is a per-token rate card from Google: each model has a USD price per 1 million input tokens, cached tokens and output tokens, with thinking tokens billed as output. Prices verified 2026-10-23 against the Gemini Developer API pricing page. On a representative task of 10,000 input tokens and 1,000 output tokens, one call costs $0.0014 on gemini-2.5-flash-lite, $0.01125 on gemini-3.8-flash, $0.03 on gemini-4-argon today and $0.06 on gemini-4-argon from December. The two promo end dates are the most important facts on this page: budgets built on October prices will be wrong by 2x within ten weeks.

How much does the Gemini API cost per million tokens?

The Gemini API price is set per model, per 1 million tokens, on the Paid tier at Standard processing. These are the text models on the Gemini Developer API:

Output costs 5x input on gemini-4-argon and the 3.8 Flash family, and 6x on gemini-3.1-pro-preview. gemini-3.8-flash and gemini-3.6-flash have identical prices, so there is no cost reason to stay on 3.6. gemini-3.5-flash-lite costs 20% more input and 67% more output than gemini-3.1-flash-lite; it is the newer model, not the cheaper one. Older Gemini 2.5 models stay cheap: gemini-2.5-flash now costs exactly the same as gemini-3.5-flash-lite, and gemini-2.5-flash-lite at $0.10 / $0.40 is the floor of the whole rate card. Audio input costs more on most models: $1.00 per million on gemini-2.5-flash. Check exact model IDs against our /models catalog before switching.

When do the Gemini 4 Argon and Flash promo prices end?

Two launch promotions end on fixed dates, and both double the price:

• gemini-4-argon is at promotional pricing through 2026-11-30. From 2026-12-01: input $2 to $4, output $10 to $20, cache reads $0.10 to $0.20. Batch, Flex and Priority double on the same day.
• gemini-3.8-flash and gemini-3.6-flash are at promotional pricing through 2026-12-31. From 2027-01-01: input $0.75 to $1.50, output $3.75 to $7.50, cache storage $0.50 to $1.00 per million tokens per hour.

On our 10,000-in / 1,000-out task, 1,000 calls cost $30 on gemini-4-argon in November and $60 in December. On gemini-3.8-flash they cost $11.25 in December and $22.50 in January. If you are benchmarking models this quarter, compare against the post-promo price, not today's. We covered the Flash cut when it was announced in Gemini 3.6 Flash's temporary 50% price cut.

How does Gemini context caching pricing work?

Context caching is Google's discount for repeated prompt prefixes; a cache hit bills those tokens at 10% of the input price on every current model. There are two kinds, per the Gemini caching docs:

• Implicit caching is on by default for Gemini 2.5 and newer. You configure nothing; savings apply when a request reuses a recent prefix. The minimum prefix is 4,096 tokens on gemini-4-argon, gemini-3.8-flash, gemini-3.6-flash and gemini-3.1-pro-preview, and 2,048 tokens on gemini-2.5-flash and gemini-2.5-pro. Cache hits appear in the response's cached token count in usage.
• Explicit caching lets you create a named cache with a TTL. You pay the cached read price plus storage per million tokens per hour: $0.50 on 3.8 Flash (promo), $1.00 on the Flash-Lite models, $4.50 on 2.5 Pro and 3.1 Pro Preview. gemini-4-argon supports implicit caching only.

Worked example: a 50,000-token system prompt on gemini-3.8-flash costs $0.0375 per call uncached and $0.00375 as a cache hit. Holding it in an explicit cache for one hour costs 0.05M x $0.50 = $0.025 in storage, so explicit caching pays for itself from the first hit in that hour. On gemini-3.1-pro-preview the same hour of storage costs $0.225, which needs about 3 hits to recover against the $0.10 uncached input cost. Put stable content (instructions, tool schemas, documents) at the start of the prompt and variable content at the end. More on the math in prompt caching break-even.

How much do Batch, Flex and Priority change the price?

Gemini has four processing tiers on the Paid plan, and the tier is a bigger lever than model choice within a family:

Flex is a preview tier that runs on "sheddable" capacity: you set service_tier to flex on a normal request, per the Flex inference docs, and accept that requests may be slower or rejected at peak. Priority is 1.8x Standard: gemini-3.8-flash Priority is $1.35 / $6.75, gemini-4-argon Priority is $3.60 / $18. A gotcha on gemini-3.1-pro-preview and gemini-2.5-pro: Batch halves input and output, but cache reads stay at the Standard price. Our comparison of OpenAI and Anthropic batch APIs applies the same "latency is free" test.

What does long context cost on Gemini Pro models?

Long-context pricing is a higher per-token rate that applies when a prompt exceeds 200,000 tokens, on gemini-3.1-pro-preview and gemini-2.5-pro. On 3.1 Pro Preview input doubles from $2 to $4 and output rises 50%, from $12 to $18. On 2.5 Pro input doubles from $1.25 to $2.50 and output rises from $10 to $15.

The cliff matters because it is priced on the prompt size of the request. A 190,000-token prompt with 2,000 output tokens on gemini-3.1-pro-preview costs $0.38 + $0.024 = $0.404. A 210,000-token prompt costs $0.84 + $0.036 = $0.876, more than double for 10% more context. If you are near the line, trim retrieval to stay under 200K or split the job. gemini-4-argon and the Flash models have no long-context surcharge on the current card. We compared this cliff with OpenAI's in GPT-5.6 long-context cliff vs Gemini 200K.

How are thinking tokens billed?

Thinking tokens are billed as output tokens at the model's output rate; Google's table labels every output price "including thinking tokens". A gemini-4-argon call that writes 300 visible tokens after 3,000 thinking tokens is billed for 3,300 output tokens: $0.033 at promo pricing, $0.066 from December. Thinking is the main reason measured Gemini bills exceed back-of-envelope estimates. Gemini 2.5 Flash introduced thinking budgets; cap them on tasks that don't need reasoning. See the reasoning token tax for worked bills.

What does a Gemini API task actually cost?

Cost per task is the number that matters for a budget, not the per-token rate. Assumptions: 10,000 input tokens (system prompt, context, question), 1,000 output tokens including thinking, no caching, Standard tier. Edit them for your workload: cost = input/1M x input price + output/1M x output price.

Two results stand out. From 2026-12-01, gemini-4-argon at $60 per 1,000 calls costs nearly twice gemini-3.1-pro-preview at $32 on this shape. And from January, gemini-3.8-flash at $22.50 costs the same as gemini-2.5-pro. Run your own token counts through the Vynaris cost calculator to compare across providers. With a 50% cache hit rate on the input, subtract 45% of the input cost; on Batch or Flex, halve the whole row.

What other charges appear on a Gemini API bill?

Tool calls and non-text modalities carry their own prices on top of tokens:

• Grounding with Google Search (Gemini 3 and newer): 5,000 free search requests per month shared across models, then $14 per 1,000 requests. One user prompt can trigger several billed searches. On Gemini 2.5: 1,500 free requests per day, then $35 per 1,000 grounded prompts.
• Grounding with Google Maps: 5,000 free per month on Gemini 3+, then $14 per 1,000.
• Code execution, URL context, computer use: no tool fee; generated code, results and fetched pages are billed as normal tokens.
• File Search: embeddings at $0.15 per 1M tokens, retrieved chunks billed as input.
• gemini-embedding-2: $0.20 per 1M text tokens.
• Image generation with gemini-nano-banana-2.1: $30 per 1M image output tokens, which is $0.0336 per 1K image, $0.0504 per 2K and $0.113 per 4K.

At $14 per 1,000, grounding adds $0.014 per search. That is more than an entire gemini-3.8-flash call on our task shape, so a grounded Flash agent can spend more on search than on tokens.

Is the Gemini API free, and how do billing tiers work?

The Gemini API has a Free tier with rate-limited access to most Flash and Flash-Lite models, but not gemini-4-argon or gemini-3.1-pro-preview. On the Free tier Google uses your content to improve its products; on the Paid tier it does not. Per the Gemini billing docs, moving to Paid means linking a Cloud billing account and prepaying at least $5 in credits.

Paid accounts move up usage tiers automatically by cumulative spend, and each tier has a billing cap and a spend-based rate limit per 10 minutes:

The spend limit is the one that surprises teams: a Tier 1 project that burns more than $10 in 10 minutes gets 429 RESOURCE_EXHAUSTED, even under its RPM and TPM limits. Rate limits apply per project, not per API key, so splitting keys does not raise them. Running a batch backfill on Standard can hit this; Batch jobs have their own enqueued-token limits instead.

Is Vertex AI priced the same as the Gemini API?

Vertex AI is Google Cloud's enterprise route to the same Gemini models, billed through a Google Cloud account instead of AI Studio prepay, and it publishes its own rate card. This page covers the Gemini Developer API only; check the Vertex AI pricing page before assuming parity. Vertex adds provisioned throughput, regional endpoints and committed-use options, and enterprise volume discounts are negotiated through Google Cloud sales. Choose Vertex when you need VPC controls, data residency or an existing Cloud commitment; choose the Gemini Developer API for the fastest start.

When is the expensive Gemini model worth it?

gemini-4-argon is worth it when a Flash model needs retries or extra agent turns to finish. After both promos end on 2027-01-01, Argon output at $20 costs 2.7x gemini-3.8-flash output at $7.50; if 3.8 Flash needs more than about 2.7x the tokens or attempts on the same task, Argon is cheaper per completed task. That happens on long-horizon coding and multi-step research. It rarely happens on extraction, classification, translation or summarization, where gemini-2.5-flash-lite at 1/50th of Argon's December output price is the right default. Measure tokens and success rate per task on your own eval set before routing by sticker price.

Where Vynaris fits

Vynaris is an OpenAI-compatible API that routes each request to the cheapest model that can handle it, across Google, OpenAI, Anthropic and open-weight providers. When Google's promo prices end, routing rules re-rate instead of your code changing. See Vynaris pricing. If every request needs Pro-level reasoning, or every request is trivial, you don't need a router; pin one model.

FAQ

How much does the Gemini API cost per 1,000 tokens?

Divide the per-million price by 1,000. gemini-3.8-flash costs $0.00075 per 1,000 input tokens and $0.00375 per 1,000 output tokens through 2026-12-31. gemini-4-argon costs $0.002 and $0.01 through 2026-11-30.

What is the cheapest Gemini API model?

gemini-2.5-flash-lite at $0.10 input / $0.40 output per 1M tokens. Batch or Flex halves that to $0.05 / $0.20.

Is the Gemini API free?

Partly. The Free tier offers rate-limited access to Flash and Flash-Lite models, and Google may use that content to improve its products. gemini-4-argon and gemini-3.1-pro-preview require the Paid tier and a $5 minimum prepay.

How much does Gemini 3 Pro cost?

The current Pro model is gemini-3.1-pro-preview: $2 input / $12 output per 1M tokens for prompts up to 200K, and $4 / $18 above that.

Does Gemini charge for thinking tokens?

Yes. Thinking tokens are billed at the output rate on every model.

How much is Gemini 2.5 Pro?

$1.25 input / $10 output per 1M tokens for prompts up to 200K tokens; $2.50 / $15 above 200K. Batch halves both.