Skip to content

Gemini API Pricing: Per-Model Cost Per Token

The real question is not which is cheapest per token, but which one can perform the tasks you need reliably, at the quality you need, for the least cost.

ModelInputOutputBatch inBatch outCache read
Gemini 3.1 Flash-Lite$0.25$1.50$0.12$0.75$0.03
Gemini 3.5 Flash-Lite$0.30$2.50$0.15$1.25$0.03
Gemini 3.8 Flash$0.75$3.75$0.38$1.88$0.07
Gemini 4 Argon$2.00$10.00$1.00$5.00$0.10

All prices in USD per 1M tokens. Batch runs asynchronously at a reduced rate; cache read applies to reused input tokens. A dash means the provider does not publish that rate.

Gemini runs from a cheap Flash-Lite model up to the Argon flagship, a roughly eight-fold spread per token. That makes the flagship look like the safe default. On most tasks it is not the choice you need.

On an everyday task, every Gemini model tied

We gave all four Gemini models the same job, pulling action items out of a short meeting note, and scored each answer against a correct version. All four scored 100. With quality level, the only thing left to separate them was cost, and there the cheapest model ran for about a seventh of the flagship.

Four Gemini models, one everyday task: same score, very different cost

Insight: On this task the cheapest Gemini was as good as the flagship, for a fraction of the cost. The flagship earns its price on harder work, not on everyday jobs. The only way to tell which kind of task you have is to run it.

Start cheap, scale up only where you need to

Three Gemini models are available on our Starter plan, and the Argon flagship on Pro. So you can start on an inexpensive model, measure it on your real task, and move up only when a task is hard enough to need it. That is the opposite of defaulting to the flagship and hoping.

Gemini input price per 1M tokens, cheapest to flagship

The headline rate is not the rate you pay

Batch requests, which run when you do not need an instant answer, bill at half the standard rate. Cached input, the tokens you resend each call (like a fixed set of instructions), bills at a small fraction of standard. Google is also running introductory prices on some Gemini models that step up on set dates, so the rate you see today is not always the rate next month. The table above tracks all of this from our live registry.

Gemini 3.1 Flash-Lite input: one headline rate, three real rates ($/1M tokens)

The best model today is not the best model forever

Google updates these models and their prices constantly, so the model that clears your bar today can slip below it next week while a rival climbs past it. Measuring once picks the right model now. Measuring on a schedule keeps quality high and cost down with no quiet regressions or cost creep.

The best model today is not the best model forever (illustrative)

Bake in great AI model hygiene

Choosing the right AI model is easy and objectively measured when you follow these steps:

  1. Define measurable quality bars for your tasks
  2. Pick the cheapest model that passes reliably
  3. Measure the whole pipeline, including RAG stores, so you’re always optimizing the right thing
  4. Monitor on a schedule so regressions and drift are fixed proactively

On Pro, your agent can run this check for you over MCP.

Agent
engineer: we default to gemini-4-argon for our extraction task. is that overkill?
> run_evaluation models: gemini-4-argon, gemini-3.1-flash-lite task: extraction bar: 90
polling ... running ... done
gemini-4-argon: quality 100 cost $0.001594/run
gemini-3.1-flash-lite: quality 100 cost $0.000220/run
gemini-3.1-flash-lite clears the bar at the same quality for 86% less. recommend switching.

What you get on Pro

  • Score any Gemini model on your own task
  • Cost, quality, and latency side by side
  • Pipeline diagnostics that identify actionable improvements
  • Scheduled benchmarks alert you when models regress or drift
  • API access to integrate measurements into your applications

Drive it from your agent, over MCP

  • Let your agent run the comparisons and read the results for you
  • Work from where you already are: your IDE, chat, or console
  • Over 40 tools your agent discovers automatically, no integration code
  • Agentic recipes for model selection, drift checks, and optimization loops

$129/mo $10 for your first week

Then $129/mo. Cancel anytime.

The table lists the Gemini models we support for evaluation, not Google's full catalogue. Rates come from our live registry and are updated frequently, including when Google's introductory prices end. The scored run is real: four models on one task, 2026-10-21. Your own task will score differently, against your prompts and data.