Does Gemini 2.5 Pro cost more for prompts over 200K tokens?

Yes. Gemini 2.5 Pro has two pricing tiers: $1.25 input / $10 output per 1M tokens for prompts up to 200,000 input tokens, and $2.50 input / $15 output per 1M once the prompt exceeds 200,000. The higher tier applies to the whole request, so a 500K-token call costs roughly double what the headline rate suggests.

Is the whole prompt billed at the higher rate, or only the tokens over 200K?

The whole prompt. Both Google and Anthropic price the entire request by its total input size — there is no marginal, per-overage tier. A 300K-token Gemini 2.5 Pro request bills all 300K at the >200K rate, not 200K at the low rate plus 100K at the high one. That is the single most common long-context budgeting mistake.

How much does Claude's 1M context window cost per token?

Claude Sonnet 4.5's 1M-context beta charges $3 input / $15 output per 1M tokens up to 200K input, then $6 input / $22.50 output above 200K — 2× input and 1.5× output for the entire request. Enter your token counts above to see the per-request and monthly figure in USD and rupees.

Why is my Gemini bill double what I estimated?

Almost always because your prompts cross the 200,000-token threshold and reprice at the long-context tier. If you estimated at the $1.25/1M headline rate but your RAG or document-analysis prompts run 300–800K tokens, every one of those calls is billed at $2.50/1M input. The premium row in this tool shows exactly how far past a naive estimate the real bill lands.

Which models have a long-context price premium and which are flat?

Among the models here, Gemini 2.5 Pro and Claude Sonnet 4.5 (1M beta) charge a premium above 200K input tokens. Gemini 2.5 Flash, GPT-4.1, and GPT-5 are flat — one price at any prompt size. That is why a flat model can undercut a tiered one at large prompt sizes; the comparison table sorts every model cheapest-first for your exact workload.

How is the tier decided — by input tokens or total tokens?

By input (prompt) tokens only. Output length does not affect which tier applies; it is priced at the tier that the input already selected. A prompt of exactly 200,000 tokens stays on the standard tier — the premium starts at 200,001. This tool uses a strict greater-than test to match the vendor boundary.

Are these prices live or up to date?

No — this is a calculator over a static price table, not a live feed. Rates are vendor list prices snapshotted on 2026-07-11. Vendors change tiers without notice, so confirm on the official pricing pages linked below before committing a budget. The table is reviewed each quarter.

How do I convert the cost to Sri Lankan rupees?

Turn on "Show LKR" and the tool multiplies the USD total by a rate you can edit — pre-filled at Rs 300, a CBSL indicative value. For an accurate figure set it to today's CBSL rate, or add the 1–3% margin your card provider charges on foreign billing.

Does this include prompt caching, batch, or reasoning-token costs?

No. This tool models the standard input/output tiered pricing only, to keep the long-context premium clear. Prompt caching and batch discounts have their own dedicated calculators, as does reasoning-token accounting. Use those alongside this one for a full picture; this page links to them under Related tools.

Developers · AI

AI Long-Context Pricing Calculator

Some models charge a premium once a prompt passes 200,000 tokens — and reprice the whole request, not just the overage. This calculator shows that jump before you ship, with the surcharge over a naive flat-rate estimate in USD and Sri Lankan rupees. No signup, sources cited below.

By Induwara Ashinsana— Executive Director, Ryzera TechnologiesUpdated Jul 11, 2026

Price your long-context requestUSD & LKR

List prices · verified Jul 2026

Model

Tiered models reprice the whole request above 200K input tokens.

Requests per month

req

How many of these calls your app makes per month.

Input tokens per request

tok

Total prompt size — the number that decides the tier.

Output tokens per request

tok

Average completion length the model returns.

Prompt size

USD → LKR rate

CBSL indicative — edit to today's rate.

Long-context tier (> threshold)500,000 input tokens > 200,000 threshold

The whole request is repriced at the long-context rate — not just the tokens over the threshold.

Per request

$1.55

Monthly cost (USD)

$1,550.00

Monthly cost (LKR)

Rs 465,000

Long-context premium

+$725.00

+87.88% vs standard rate

Per-request breakdown

Component	Tokens	Applied rate /1M	Cost
Input	500,000	$2.50$1.25	$1.25
Output	20,000	$15.00$10.00	$0.30
Per-request total			$1.55
Monthly total (1,000 req)Rs 465,000			$1,550.00
Premium vs a naive standard-rate estimate			+$725.00

Tier is decided by input tokens alone. A naive flat-rate estimate at the standard price would predict $825.00/month — the row above is how far the real bill lands beyond it.

Same workload, every model (cheapest first)

Cheapest option saves $1,350.00/mo (Rs 405,000)

Model	Tier	Monthly USD	Monthly LKR	vs selected
Google	flat	$200.00	Rs 60,000	-87.1%
OpenAI	flat	$825.00	Rs 247,500	-46.77%
OpenAI	flat	$1,160.00	Rs 348,000	-25.16%
Google	long	$1,550.00	Rs 465,000	—
Anthropic	long	$3,450.00	Rs 1,035,000	+122.58%

Tap any row to recompute the headline for that model. A flat-rate model can beat a tiered one once your prompt crosses 200K tokens — that flip is the whole point.

Vendor list prices snapshotted on 2026-07-11 — not a live feed.

How it works

A handful of long-context models — Gemini 2.5 Pro and the Claude Sonnet 4.5 1M-context beta among them — bill in two tiers. Below a 200,000-input-token threshold you pay the standard rate; above it, a higher long-context rate. The trap is that both Google and Anthropic apply the higher rate to the entire request, not only to the tokens beyond the threshold. There is no marginal, per-overage tier for long context at any listed vendor, so a prompt one token over the line reprices in full.

For a model with standard input/output rates stdIn / stdOut and long-context rates longIn / longOut (USD per 1M tokens), the calculator runs:

Tier selection by input tokens: tier = inputTokens > 200,000 ? long : standard. The test is strict — exactly 200,000 tokens stays on the standard tier. Flat models skip this step entirely.
Rate pick: inRate = long ? longIn : stdIn and likewise for output. The long tier changes both input and output pricing.
Per-request cost: costIn = inputTokens ÷ 1e6 × inRate, costOut = outputTokens ÷ 1e6 × outRate, summed.
Monthly = per-request × requests per month. LKR = USD × your editable CBSL rate.
Premium: the same tokens are priced again at the standard rate, and the difference is the surcharge — surcharge = perReq − stdPerReq, as a percentage of the standard cost. It is $0 for standard-tier and flat models.

Every per-1M figure in the underlying data module carries an inline source URL, and the arithmetic is reconciled against an independent per-1,000-token formula so it can't drift. The comparison table reruns the whole calculation for every model on each input change and sorts cheapest-first — because a flat-rate model often undercuts a tiered one once the prompt is large enough.

Worked examples

RAG app on Gemini 2.5 Pro — the classic surprise

500K in · 20K out · 1,000 req/mo · std $1.25/$10, long $2.50/$15

500,000 > 200,000 → long-context tier (whole request reprices)
costIn = 500,000 ÷ 1e6 × $2.50 = $1.25
costOut = 20,000 ÷ 1e6 × $15 = $0.30
perReq = $1.55 → monthly = $1.55 × 1,000 = $1,550.00
naive std estimate = 0.5×$1.25 + 0.02×$10 = $0.825/req = $825.00/mo
premium = $1,550 − $825 = +$725.00/mo (+87.88%)

Document analysis on Claude Sonnet 4.5 (1M beta)

300K in · 5K out · 500 req/mo · std $3/$15, long $6/$22.50

300,000 > 200,000 → long-context tier
costIn = 300,000 ÷ 1e6 × $6 = $1.80
costOut = 5,000 ÷ 1e6 × $22.50 = $0.1125
perReq = $1.9125 → monthly = $1.9125 × 500 = $956.25
naive std estimate = 0.3×$3 + 0.005×$15 = $0.975/req
premium = $1.9125 − $0.975 = +$0.9375/req (+96.15%)

Below threshold — no premium (boundary case)

Gemini 2.5 Pro · 150K in · 10K out · 1,000 req/mo

150,000 ≤ 200,000 → standard tier, no premium
costIn = 150,000 ÷ 1e6 × $1.25 = $0.1875
costOut = 10,000 ÷ 1e6 × $10 = $0.10
perReq = $0.2875 → monthly = $287.50
premium = $0.00 (0%) — safely under the line
Note: exactly 200,000 tokens also stays standard; the jump is at 200,001.

Frequently asked questions

Sources & references

Prices were last cross-checked against these sources on 2026-07-11. This tool is a calculator over a snapshotted price table, not a live feed — confirm the current rate on the official page before committing a budget.

Related tools

LiveAI

AI API Cost Calculator

Estimate the monthly and per-request USD and LKR bill for any major LLM API. Pick a model, enter input/output tokens and requests per month, and compare every model cheapest-first — with optional 50% batch and cached-input discounts.

Open tool

LiveAI

Reasoning Token Cost Calc

Estimate the true cost of reasoning-model API calls by accounting for the hidden reasoning/thinking tokens that o-series, Claude, and Gemini bill at output rates. See per-call and monthly cost in USD and LKR, plus how much more it is than a naive estimate.

Open tool

LiveAI

LLM Price History Tracker

Track the dated API price of every major large-language model — GPT, Claude, Gemini, Grok and hosted open-weights — from launch to today, in one filterable table and trend chart. A cost time-machine shows what your monthly token workload would have cost in 2023, 2024, 2025 and now, and exactly how much you save today. Every figure is a cited, dated list price.

Open tool

Rate this tool

Be the first to rate

Comments & feedback

Spotted a bug or want an improvement? Tell us — our team reviews every comment, and good ideas get built. Comments are public and anonymous.

Spotted a stale price, a new tier, or want another model added?

Email me at [email protected] — most fixes ship within 24 hours.