Gemini API Pricing After the December 2026 Increase

Gemini 4 Argon costs $2 input / $10 output per 1M tokens through November 30, 2026, and $4 / $20 from December 1, 2026. Gemini 3.8 Flash follows a month later: $0.75 / $3.75 through December 31, then $1.50 / $7.50 from January 1, 2027. Nano Banana 2.1 image pricing has no scheduled change. All three dates come from Google's Gemini API pricing page.

On Orux AI, Gemini 4 Argon is 132 / 660 Credits per 1M tokens ($0.66 / $3.30) and Gemini 3.8 Flash is 42 / 210 Credits ($0.21 / $1.05), as published on Oct 29, 2026. The tables below show every mode Google sells, what the increase does to a monthly bill, and the details that change the real cost: thinking tokens, caching and the free tier.

What changes and when

Swipe horizontally to view the full table.

Google's page lists dates, not a time of day, so if you run cost-sensitive jobs around the switch, assume the new rate could apply from the start of December 1 in any time zone. The page shows no exception for existing users, so plan for the new rate on all paid usage.

Gemini 4 Argon pricing

Per 1M tokens. Google rates from the Gemini API pricing page; Orux AI rate as of Oct 29, 2026 (1 Credit = $0.005):

Swipe horizontally to view the full table.

Today the Orux AI rate is 33% of Google's Standard price. Three facts from Google's page matter for budgeting:

• Argon is not available on Google's free tier.
• There is a single rate regardless of prompt length; Argon has no long-context surcharge.
• Thinking tokens are counted and billed as output. A short visible answer can still carry a long, billed reasoning trace.

Orux AI does not offer batch processing, so the Batch and Flex rows are only available direct from Google. As of Oct 29, Orux AI's standard rate for Argon is also below Google's current batch rate.

Gemini 3.8 Flash pricing

Swipe horizontally to view the full table.

Google also charges $0.50 per 1M tokens per hour to store a cached context, rising to $1.00 on January 1. Unlike Argon, 3.8 Flash does have a free tier on Google's Standard mode, which is useful for prototyping but comes with rate limits.

The Orux AI rate is 28% of Google's current Standard price. Like Argon, Flash bills thinking tokens as output, and there is no long-context tier.

Nano Banana 2.1 image pricing

Nano Banana 2.1 is Google's image model in the Gemini API. Google bills it by tokens: $30 per 1M image-output tokens, which Google itself converts into per-image prices. Input text and images cost $1.50 per 1M tokens on top, and any text or thinking output costs $7.50 per 1M.

Swipe horizontally to view the full table.

Two honest caveats. Google's batch price is lower than Orux AI at 1K and 2K, by about 7%, so if you generate large offline batches at those sizes and can wait, going direct is cheaper. And Google's per-image figure excludes input tokens and any thinking output, whereas Orux AI charges one flat price per successful image at each resolution, with reference images included.

There is no scheduled price change for Nano Banana 2.1. Our Nano Banana 2.1 API guide covers parameters, editing with reference images and a working script.

Monthly cost before vs after

Three usage levels, all with a 5:1 input-to-output ratio and no caching. Google figures use Standard mode.

Swipe horizontally to view the full table.

The Orux AI columns use the rates published on Oct 29, 2026. Orux AI has not announced a change tied to Google's dates; if one happens it will appear on the model pages.

What the table means in practice: a team spending $400 a month on Argon today will see $800 in December with no change in traffic. Three levers help:

1. Cache aggressively. Argon's cache read is 5% of the input price on Google, before and after the increase.
2. Move tolerant traffic to 3.8 Flash. Even after January 1, Flash Standard at $1.50 / $7.50 is well below Argon's new $4 / $20.
3. Cap thinking. Because reasoning tokens are billed as output, the output column is usually where the bill grows fastest. Set a sensible max_tokens, measure completion_tokens, and do not leave the ceiling at the maximum for simple tasks.

Thinking tokens and max_tokens

Google bills thinking tokens at the output rate on both Argon and 3.8 Flash. On Orux AI, the Gemini model pages note that max_tokens limits response and reasoning tokens together. That has two consequences.

If max_tokens is too small, a request can spend its whole budget thinking and return a short or empty answer, which you still pay for. If it is too large on a simple prompt, you pay for reasoning you did not need. A practical approach is to start at around 2,048 for normal tasks, log completion_tokens for a day of real traffic, then tighten.

Should you switch models before December 1?

Not automatically. A price increase is a reason to measure, not a reason to migrate. Before you touch production, answer three questions with your own logs.

How much of your Argon traffic actually needs Argon? Google positions Argon for complex coding, agentic, math and science work. Classification, extraction, short summaries and routing rarely need it. Send a sample of a few hundred real requests to 3.8 Flash and compare the answers. If most of them are acceptable, routing those requests to Flash cuts their cost by more than half compared with Argon's December price, even after Flash's own increase in January.

Where does your output go? Look at completion_tokens against the length of the visible answer. If reasoning dominates, a tighter max_tokens or a simpler prompt may save more than any model change.

How much of your input repeats? System prompts, tool definitions and retrieved documents that are identical across requests are candidates for caching. On Google, a cached Argon read stays at 5% of the input price after December 1.

If you decide to compare Gemini against other vendors at the same time, keep the test identical: same prompts, same max_tokens, same scoring. Switching the model string on an OpenAI-compatible endpoint makes that a one-line change, which is the main reason to keep vendor-specific code out of the request path.

Call Gemini via an OpenAI-compatible endpoint

Orux AI serves both Gemini chat models through https://orux.top/api/v1/chat/completions, so the OpenAI Python SDK works with base_url changed and the model name set to the Orux alias. The script below streams an Argon response, then projects what the same monthly volume would cost at each rate.

import os
from openai import OpenAI

client = OpenAI(base_url="https://orux.top/api/v1", api_key=os.environ["ORUX_API_KEY"])

stream = client.chat.completions.create(
    model="gemini-4-argon",          # or "gemini-3.8-flash"
    messages=[{"role": "user", "content": "List three risks of migrating a cron job to a queue."}],
    max_tokens=2048,                 # covers reasoning + visible answer
    stream=True,
    stream_options={"include_usage": True},
)
usage = None
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)
    if chunk.usage:
        usage = chunk.usage
print()

# USD per 1M tokens (input, output). Orux AI rates converted at 1 Credit = $0.005.
RATES = {
    "Google Argon until Nov 30": (2.00, 10.00),
    "Google Argon from Dec 1":   (4.00, 20.00),
    "Orux AI Argon (Oct 29)":    (0.66, 3.30),
    "Google Flash until Dec 31": (0.75, 3.75),
    "Google Flash from Jan 1":   (1.50, 7.50),
    "Orux AI Flash (Oct 29)":    (0.21, 1.05),
}

def monthly(in_m: float, out_m: float) -> None:
    """in_m / out_m: millions of tokens per month."""
    for name, (i, o) in RATES.items():
        print(f"{name:28s} ${in_m * i + out_m * o:,.2f}")

if usage:
    print(f"this call: {usage.prompt_tokens} in / {usage.completion_tokens} out")
monthly(100, 20)

Whether stream_options usage arrives depends on the client and model; if usage stays empty, make a non-streaming call to read token counts. To compare Argon with Claude Sonnet 5.5 and GPT-6 Luna on one prompt, change the model string; our three-way comparison does exactly that, and the Claude API pricing guide has the equivalent tables for Anthropic.

FAQ

Q.When does Gemini 4 Argon get more expensive?

On December 1, 2026. Google's pricing page lists $2 input and $10 output per 1M tokens through November 30, then $4 and $20 from December 1. Batch, Flex and Priority rates double on the same date, and cached reads go from $0.10 to $0.20.

Q.Is there a free tier for the Gemini API?

Gemini 3.8 Flash has a free tier on Google's Standard mode, with rate limits. Gemini 4 Argon and Nano Banana 2.1 are not available on the free tier. Orux AI has no free tier or subscription: you top up a prepaid balance and pay per request, and failed requests are not charged.

Q.Are thinking tokens billed?

Yes. Google counts thinking tokens and bills them at the output rate on Gemini 4 Argon and 3.8 Flash. On Orux AI, max_tokens caps reasoning and the visible answer together, so set it deliberately and watch completion_tokens in the response usage.

Q.How much is Nano Banana 2.1 per image?

Google lists $0.0336 at 1K, $0.0504 at 2K and $0.113 at 4K on Standard, and half that on Batch, excluding input tokens. On Orux AI it is 3.6, 5.4 and 8.1 Credits ($0.018, $0.027, $0.0405) per successful image, as of Oct 29, 2026.

Q.Will Orux AI prices change on December 1?

Orux AI has not announced a change linked to Google's increase. The rates in this article were published on Oct 29, 2026. Current prices are always shown on each model page, and any change will be posted there, so check the page before planning a large job.

Sources and price check

Prices checked on 2026-10-29. Google rates, price-change dates, free-tier availability and thinking-token billing: Gemini API pricing. Release and routing dates: Gemini API changelog. Orux AI rates: Gemini 4 Argon, Gemini 3.8 Flash and Nano Banana 2.1.