Gemini API Pricing in 2026: Text, Video and Speech Compared

Google prices Gemini three different ways depending on what you're generating: per million tokens for text, per clip for video, and per call for speech. Mixing those up is the fastest way to get a cost model wrong.

This post lists all three, with the direct price beside ours, then shows where the money actually goes.

All prices USD, checked 15 August 2026 (1 credit = $0.01). Percentages are against standard rates; the $1,250 top-up pack (+10% bonus) adds a few more points.

Gemini text pricing, per million tokens

Model pages: Gemini 3.1 Pro · Gemini 3.6 Flash · Gemini 3 Flash · Gemini 2.5 Pro · Gemini 2.5 Flash.

The interesting line here isn't the discount, it's the spread inside the family. Gemini 2.5 Flash output is $1.3117 per million; Gemini 3.1 Pro output is $6.2215 — 4.7× more. Most products send everything to the Pro tier by default and pay nearly five times too much for work a Flash model handles fine.

Worth knowing if you're picking on price alone: GPT-5.6 Terra is now $0.65 in / $3.90 out, which undercuts Gemini 3.1 Pro on both sides. If your workload isn't tied to Gemini specifically, benchmark Terra before you commit.

Gemini video pricing, per clip

Gemini Omni is billed per finished clip, not per second — so cost scales with duration in fixed steps.

Two things fall out of per-clip billing:

• 1080p costs almost nothing extra at short durations — 2 cents more at 4 seconds, and identical at 6 seconds. Render high.
• Duration is your real cost dial. If your product shows short loops, generate short.

Google's own Veo 3.1 is on the same key at $1.7286 per 720p clip (−46%), and Veo 3.1 Fast at $0.5847 (−51%) — three times cheaper than standard Veo for a lot of use cases. For per-second models, Seedance 2.0 Mini starts at $0.019/s and Grok Imagine 1.5 at $0.0268/s — the two cheapest video options on the platform, both around 74–78% below direct. Browse them all under video models.

Gemini speech pricing

Model page: Gemini 3.1 Flash TTS.

What a real workload costs

A document-summarisation feature: 8,000 input tokens and 500 output tokens per document, 100,000 documents a month — 800M input, 50M output.

Note the bigger lesson in that table: choosing Flash over Pro saves $899 a month at our prices — five times more than the discount does on Flash. Model selection beats vendor selection. The discount is on top.

How to call Gemini

POST https://you.bot/api/v1/generate
Authorization: Bearer YOUR_API_KEY

{
  "modelId": "gemini-3-1-pro",
  "input": { "prompt": "Summarise this support thread in three bullets." }
}

Text models return the answer inline in the create response — nothing to poll. Video is asynchronous, so you poll or pass a callbackUrl:

POST https://you.bot/api/v1/generate
{ "modelId": "gemini-omni-video", "input": { "prompt": "...", "duration": 4 } }

GET https://you.bot/api/v1/task/{taskId}?model=gemini-omni-video

Swapping between Gemini, Claude and GPT is a one-string change, which makes it cheap to route easy requests to a Flash-class model and hard ones to a Pro-class model. More on that in One key, every model.

Three ways to cut a Gemini bill

Route by difficulty. Classification, extraction and routing go to Gemini 2.5 Flash or 3 Flash. Reserve 3.1 Pro for genuinely hard reasoning. This is the single biggest lever.

Cap output. Output is 5–8× input across the family. Asking for JSON instead of prose, and capping max_tokens at what you'll render, cuts more than any vendor switch.

Don't pay for failures. Failed, errored and empty generations are refunded automatically here, which matters most on video where one clip costs more than a thousand text calls. (How that works.)

Comparing across vendors as well? The cheapest AI APIs in 2026 puts Gemini beside GPT, Claude and the image and video models, and Claude API pricing does the same for Anthropic's line-up.

FAQ

How much does the Gemini API cost? Direct, Gemini 2.5 Flash is $0.30 per million input tokens and $2.50 output; Gemini 3.1 Pro is $2.00/$12.00. Here the same models are $0.1574/$1.3117 and $0.9745/$6.2215 — 47–51% less at standard rates, checked 15 August 2026.

Which Gemini model is cheapest? Gemini 2.5 Flash, at $0.1574 per million input tokens and $1.3117 output. It's roughly 4.7× cheaper on output than Gemini 3.1 Pro.

How is Gemini video billed? Gemini Omni bills per clip, not per second: $0.31 for a 4-second 720p clip, $0.41 at 6 seconds, $0.73 for 4 seconds at 4K. 1080p costs almost nothing extra at short durations.

Is Gemini cheaper than Claude or GPT? At the Flash tiers, yes — Gemini 2.5 Flash output is $1.3117 per million versus $7.702 for Claude Sonnet 5. At the Pro tier, GPT-5.6 Terra is now cheaper than Gemini 3.1 Pro on both input and output. Pick on capability for your task, then optimise price.

Do I pay for failed generations? No — failed, errored and empty generations are refunded automatically.