AI Translation Cost Calculator

Every rate below is a public list price per 1,000,000 characters of source text. One million characters is roughly 166,667 English words, or about 330 A4 pages of prose. Rupee figures use the page default of Rs 305/USD; override the rate in the calculator if your card or bank settles at a different one.

DeepL API Free has no metered rate because it cannot be billed: it serves up to 500,000 characters a month and then stops. The GPT-4o mini row is converted from token pricing at roughly four characters per token and is an estimate, not a quoted rate.

Every major translation API bills on characters of source text, including spaces, not words and not the translated output. This calculator converts whatever you enter (pasted text, a word count, or a character count) into a character count, then applies each provider's published rate and free allowance. The method is deterministic and runs entirely in your browser; no text is sent anywhere.

Step 1, get to a character count

Pasted text is counted by Unicode code points, so Sinhala, Tamil, and emoji count exactly as the APIs bill them, a Sinhala character is one billable character, not one byte. A word count is converted at 6 characters per word (five letters plus one space), the standard typography assumption. If you only have a document and no count, run it through the character counter or the word counter first and paste the number here. Counting the real text always beats estimating from the word count, because the 6-chars-per-word average understates Sinhala and Tamil.

Step 2, multiply by languages and runs

Each target language is a separate billable pass over the whole source text; there is no discount for translating the same paragraph twice. Translations per month covers content you re-translate on a schedule, a product feed, a notice board, a news page:

monthlyChars = chars × targetLanguages × translationsPerMonth

Step 3, subtract the free tier once

Free allowances are monthly, not per run, so the subtraction happens once against the monthly total rather than against each pass. When the free tier is applied: Google 500,000; Amazon 2,000,000 (first 12 months only); Azure F0 2,000,000; DeepL Free 500,000as a hard cap, above it the Free plan is marked unavailable rather than billed, because DeepL does not charge overage on it. DeepL Pro has no free allowance and bills from the first character. Boundaries are inclusive: exactly 500,000 characters is still free on Google, and 500,001 is not. Clear the toggle if you have already spent the month's allowance on another project, the calculator then bills every character and marks DeepL Free unusable.

Step 4, the per-provider formulas

Billable characters (B) are divided by 1,000,000 and multiplied by the per-million rate, plus any fixed monthly base fee:

• Google: B ÷ 1,000,000 × $20
• Amazon (Standard): B ÷ 1,000,000 × $15
• Azure S1: B ÷ 1,000,000 × $10
• DeepL Pro: $5.49 base + B ÷ 1,000,000 × $25
• GPT-4o mini (est.): tokens = chars ÷ 4; cost = tokens ÷ 1,000,000 × ($0.15 in + $0.60 out)

Each usage figure is then recomputed a second way, as a strict per-character rate (rate ÷ 1,000,000 × characters) rather than by dividing characters by a million, and the two paths must agree to the cent. That cross-check is what generates the quick-reference table above, which is why the table and the tool can never disagree.

Step 5, dollars into rupees

The USD subtotal is multiplied by the USD→LKR rate, defaulting to Rs 305, a Central Bank of Sri Lanka indicative figure you can override. Cloud providers bill in dollars and your bank converts at its own selling rate plus any card fee, so the honest planning number is a few rupees above the indicative rate. Check today's LKR exchange rate and type the rate your bank actually gives you.

Step 6, read the answer as volume bands

Because three of the five plans lead with a free allowance, the cheapest provider moves with your volume instead of staying fixed:

• Under 500,000 characters a month, free on Google, Azure F0, Amazon and DeepL Free alike. Choose on output quality and language coverage, because price is not deciding anything.
• 500,000 to 2,000,000, Azure F0 and Amazon's first-year tier still cost nothing, Google starts billing at character 500,001, and DeepL Free stops serving rather than billing.
• Above 2,000,000, the allowances are spent and list price decides: Azure S1 at $10 per million, then Amazon at $15, Google at $20, and DeepL Pro at $25 plus its $5.49 monthly fee.
• From month 13, Amazon's free tier expires and its bill jumps at unchanged volume, while Azure F0's 2,000,000 a month does not expire. The last worked example below prices that cliff.

DeepL Pro is the dearest row on every one of those lines, which is the point of it: you are paying for output, glossaries and the document endpoint rather than for characters. When features weigh as much as price, the translation API comparison ranks the same providers on language coverage, glossary support and commercial-use terms instead of cost alone.

Why the GPT row is an estimate

GPT-4o mini is the only token-billed entry. Tokens are not characters: the o200k_base tokenizer averages about four characters per token for English prose, but far fewer for Sinhala and Tamil, where a single syllable can cost several tokens. The row also assumes the translated output roughly mirrors the input length, since you pay for both directions. If token accuracy matters to your budget, count the real thing with the AI token counter and treat this row as an order-of-magnitude comparison only.

What this model leaves out

All figures are list prices for standard text translation. Out of scope: volume and committed-use discounts, custom or adaptive model surcharges, document-translation endpoints (which price per page or per document, not per character), glossary and formality features, batch storage and egress, and taxes. Retries count, a failed request that you resend is billed twice by every provider, so add a margin if your pipeline is not idempotent.

The price gap is not close. A 1,000-word page is about 6,000 characters, which Azure will translate for $0.06 — under Rs 20. A professional translator quoting the low end of published market rates, USD 0.08 a word, charges USD 80 for the same page: about 1,300 times more. That ratio decides the workflow rather than the question, and the usual answer is to machine-translate everything and spend the human budget on post-editing what matters.

Where machine output is enough: internal documentation, product descriptions, support-article first drafts, user-generated content, search and discovery text, and anything a reader would otherwise not see in their language at all. You can try the quality for your own language pair free with the AI translator before you commit to a paid key.

Where it is not enough, and a Sri Lankan context matters: anything filed with a government department, court documents, affidavits, educational certificates and marriage or birth records generally need a sworn or certified translation that a machine cannot provide, whatever the quality. Medical instructions, legal terms and contracts carry real liability if a term is rendered loosely. Marketing copy and brand names usually need a human because the failure mode is embarrassment, not mistranslation. A practical middle path for Sinhala and Tamil work is machine translation plus a paid human review pass. You still pay per word for the review, but you pay it on a draft rather than a blank page.

If translation is one line in a larger AI budget, the same per-million-unit method applies to the rest of the stack: the transcription cost calculator prices speech-to-text per minute of audio, and the AI API cost calculator prices general LLM calls per million tokens.

• Google Cloud Translation: pricing ($20/1M, 500K free)
• DeepL API: Free & Pro plan pricing ($5.49/mo + $25/1M)
• Amazon Translate: pricing ($15/1M, 2M free for 12 months)
• Microsoft Azure AI Translator: pricing (S1 $10/1M, F0 2M free)
• OpenAI API: GPT-4o mini pricing ($0.15 in / $0.60 out per 1M tokens)
• Central Bank of Sri Lanka: indicative USD/LKR exchange rate

All provider rates and free allowances were last cross-checked against the official pricing pages on 2026-06-05. Human translation rates quoted on this page are market ranges for context, not figures this tool computes. Translation-API pricing changes occasionally; if a figure looks off, email me and I'll re-verify within 24 hours.