gemini 3 pro api pricing per million tokens

Gemini 3 Pro Preview is no longer a current Google API purchase: Google's model page says it shut down on March 9, 2026. For new deployments, distinguish that retired model from Gemini 3.1 Pro, whose displayed base-tier reference rates are $2 input and$12 output per million tokens as of 2026-10.

Crazyrouter is an AI API gateway that advertises 300+ models, documents OpenAI-, Anthropic-, and Gemini-compatible endpoints, and lists Gemini 3.1 Pro input at $1.10 per 1M tokens.

What is Gemini 3 Pro, and why does the version matter?#

Gemini 3 Pro Preview used the Google identifier gemini-3-pro-preview. Google's model status page now labels it shut down and directs developers to Gemini 3.1 Pro Preview. A search result containing the older name may therefore describe historical pricing, not an available service.

Gemini 3.1 Pro is a distinct model version. Its price row is useful for planning a migration, but it must not be relabeled as a verified current quote for Gemini 3 Pro. Similarly, Gemini 3 Pro Image is an image-generation product; matching part of the name does not make its image billing relevant to ordinary text generation.

The safe first step is to copy the model ID from your existing request or invoice. Then verify that identifier's current availability and billing route before comparing dollar amounts. A friendly display name in an application can hide an alias or a version change.

gemini 3 pro api pricing per million tokens#

There is no verified current Gemini 3 Pro quote on the dedicated Crazyrouter price-page path checked for this article. Rather than invent a number for a retired Google preview, the following table explicitly compares the current Gemini 3.1 Pro base tier shown on the Gemini 3.1 Pro pricing page.

All numerical prices in this article come from that Crazyrouter model price page, checked October 13 with an October 12 snapshot. Provider reference figures can lag changes; the linked official source remains authoritative for contractual billing. “Not verified” does not mean unavailable or expensive—it means this comparison lacks a supported like-for-like number.

When does the longer-context tier apply?#

The same pricing page lists Google AI reference tiers of $2 input and$12 output per million for requests with at most 200,000 input tokens. Above 200,000 input tokens, it lists $4 input and$18 output per million.

This threshold applies per request, not to your monthly token total. Many small requests totaling millions of tokens do not become one long-context request. Conversely, a single large prompt can require the higher tier even in a low-volume project.

Do not extrapolate Crazyrouter's base-tier percentage difference to an unpublished long-context gateway quote. Check the actual billing rule for the route you select, especially when adding a long system prompt, conversation history, retrieved documents, or multimodal input.

How much would a real workload cost?#

For an illustrative workload of one million uncached input tokens and 100,000 output tokens, spread across requests that remain inside the base tier, the displayed standard Google reference rates imply $3.20. The displayed Crazyrouter rates imply$1.76 for the same counted tokens before any extras.

The arithmetic is input tokens divided by one million times the input rate, plus output tokens divided by one million times the output rate. This example is calculated, not a benchmark or a promise about your invoice. It excludes retries, tools, taxes, cache storage, and other separately billed services.

Inspect usage fields rather than counting only text visible on screen. Thinking output and tool activity can affect charges; caching has its own eligibility and pricing. A consumer Gemini subscription is a different product and should not be used as the denominator in an API token comparison.

How to call the current model instead of the retired preview#

The following gateway examples use the catalog identifier gemini-3.1-pro and the OpenAI-compatible base URL. This is a migration example for a different model, not a claim that gemini-3-pro-preview still works. Store your API secret in CRAZYROUTER_API_KEY and install the OpenAI SDK for Python or Node.js.

Python#

import os
from openai import OpenAI
client = OpenAI(
    api_key=os.environ["CRAZYROUTER_API_KEY"],
    base_url="https://api.crazyrouter.com/v1",
)
result = client.chat.completions.create(
    model="gemini-3.1-pro",
    messages=[{"role": "user", "content": "Reply with exactly: pricing checked"}],
)
print(result.choices[0].message.content)

Node.js#

import OpenAI from "openai";
const client = new OpenAI({
  apiKey: process.env.CRAZYROUTER_API_KEY,
  baseURL: "https://api.crazyrouter.com/v1",
});
const result = await client.chat.completions.create({
  model: "gemini-3.1-pro",
  messages: [{ role: "user", content: "Reply with exactly: pricing checked" }],
});
console.log(result.choices[0].message.content);

cURL#

curl --fail-with-body https://api.crazyrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $CRAZYROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gemini-3.1-pro","messages":[{"role":"user","content":"Reply with exactly: pricing checked"}]}'

Tested on 2026-10-13: all three examples returned pricing checked. The cURL response identified the routed model as gemini-3.1-pro-preview, so the gateway catalog name should not be mistaken for a different stable Google model. This was a basic text smoke test, not a full compatibility benchmark.

Before migrating production traffic, compare a fixed set of your own prompts and confirm tool calls, output format, and streaming behavior. The Google model catalog and full pricing list help identify alternatives; neither replaces testing your actual account permissions.

FAQ#

Is Gemini 3 Pro Preview still available on Google's API?#

Google's documentation marks it shut down on March 9, 2026. Do not build a new integration around that old identifier without checking the exact service and its current status.

Are Gemini 3 Pro and Gemini 3.1 Pro prices interchangeable?#

No, they name different versions. This article quotes the current Gemini 3.1 Pro row explicitly and does not present it as a live offer for the retired preview.

Is Crazyrouter cheaper than standard Gemini 3.1 Pro pricing in 2026?#

The displayed base-tier rates are $1.10/$6.60 versus the standard reference $2/$12 per million input/output tokens. Batch, cached, cloud-contract, and longer-context billing require separate comparisons.

Does one million tokens mean one million words?#

No: tokens are model-specific units of encoded input and output. Read the request's usage counters, and remember that input and generated output have separate rates.

Can I infer API access from a Gemini app subscription?#

An app subscription is not evidence of developer API billing entitlement. Check the account, product, and endpoint that will execute your request, then budget against its terms.

Summary#

Confirm the model version before the price: a retired preview, a newer Pro model, and an image model are not interchangeable. Review Google's official pricing, then create a gateway account if you want to test a supported route with a small, measured workload.

Related on crazyrouter.com#

• Browse models by vendor
• OpenRouter alternatives