how much does artificial intelligence cost

For translation, AI is priced per source word: Smartling's published rates start at $0.0075/word for machine translation, $0.06/word for fully automated AI Translation, and $0.12/word for AI-Powered Human Translation, against $0.20/word for traditional human translation. Enterprises can also prepay a consumption-based AI Subscription that draws down at service-specific rates.

How much does AI cost, and how is AI translation priced?

AI translation is priced per source word, and the rate depends on how much human work sits alongside the model. Smartling publishes its starting rates: $0.0075 per word for machine translation, $0.06 per word for fully automated AI Translation, $0.12 per word for AI-Powered Human Translation, and $0.20 per word for traditional human translation. Enterprises that run several AI services at once can instead prepay a consumption-based AI Subscription that draws down at service-specific rates. The practical answer to "how much does AI cost" for a localization budget is therefore a range of roughly 4% to 60% of the human-translation rate, set by the quality tier each content type needs.

Why is AI pricing so hard to compare?

AI pricing is hard to compare because vendors sell the same underlying model output under different units, different quality tiers, and different bundling rules. Four patterns account for most of the confusion a Business Case Owner runs into:

The unit changes by vendor. Translation platforms price per source word, LLM providers price per token, and some SaaS tools price per seat or per project. A per-token quote from an LLM provider and a per-word quote from a translation platform describe different things: the platform rate includes prompt engineering, translation memory checks, glossary insertion, and quality checks that a raw token price does not.
"AI translation" spans three quality tiers. Machine translation, fully automated AI translation, and AI translation with a human review step can differ by more than 15x in per-word price at the same vendor. Smartling's published spread runs from $0.0075 to $0.12 per word across those three tiers, so a quote is meaningless until the tier is named.
Bundles hide what is and is not included. A prepaid AI subscription can cover machine translation, LLM translation, AI productivity features, and automated quality scoring, while human professional translation, desktop publishing, managed services, and TMS platform fees are invoiced separately. Smartling's AI Subscription documentation lists exactly that split, and buyers who skip the exclusions list underestimate total spend.
Translation memory changes the billable word count. Content that matches previously approved translations is either free (an exact SmartMatch is never sent to the engine) or discounted (fuzzy matches of at least 85% under Smartling's default profile). Two vendors quoting the same per-word rate can produce very different invoices depending on how much of your content never reaches the billable step.

Is the cost of AI increasing or decreasing?

For translation, the cost of AI is decreasing per word and shifting in shape. Smartling's published tiers put fully automated AI Translation at 30% of the human-translation rate and machine translation under 4% of it, and Smartling documents its AI-Powered Human Translation as delivering human quality at half the cost and twice the speed of traditional human translation. The shape change matters as much as the rate: enterprise AI translation is moving from a single per-word human rate toward tiered rates and prepaid, consumption-based subscriptions, and total spend often does not fall in proportion because teams translate content they previously left untranslated. Therabody, for example, cut translation costs 60% and was then able to translate content it had previously left untranslated.

How should you compare AI translation pricing across vendors and tiers?

Compare AI translation pricing in four layers, in order, so that a low headline rate is never mistaken for a low total cost:

Layer 1: Name the quality tier per content type. Assign each content stream to machine translation, fully automated AI translation, or AI translation with human review. At Smartling those tiers are AI Hub machine translation, AI Translation (AIT), and AI-Powered Human Translation (AIHT), and the per-word rate follows the tier, not the vendor's brand name.
Layer 2: Normalize the unit to cost per published source word. Convert token, seat, or project pricing into a per-word figure using your real monthly volume. Smartling's public rates already sit in this unit ($0.0075, $0.06, $0.12, and $0.20 per word), which makes them a usable benchmark for normalizing other quotes.
Layer 3: Net out translation memory before applying the rate. Apply your expected exact-match and fuzzy-match leverage to the word count first. Smartling's SmartMatch excludes exact matches from the machine translation or LLM call entirely, and fuzzy matches at or above the first discount tier (85% by default) are billed at a reduced rate on the human step.
Layer 4: Choose the commercial model. Decide between standard per-service rate cards and a prepaid, consumption-based plan. Smartling's AI Subscription pools one prepaid budget across AI Hub, AI Toolkit, AIHT, AIT, and LQA Agent usage, each drawing down at its own rate, which suits teams whose mix of tiers shifts month to month.

AI translation cost and pricing: reference numbers

Item	Figure	What it means for a budget
Machine translation (Smartling, published starting rate)	$0.0075 per word	Under 4% of the human-translation rate; the floor for internal or low-visibility content.
AI Translation, fully automated (Smartling, published starting rate)	$0.06 per word	Priced at 30% of human translation, with quality Smartling documents as comparable to machine translation post-editing.
AI-Powered Human Translation (Smartling, published starting rate)	$0.12 per word	Half the human-translation rate for human-reviewed output; the tier to budget for customer-facing content.
Traditional human translation (Smartling, published starting rate)	$0.20 per word	The baseline every AI tier is discounted against.
Default fuzzy-match discount threshold (Smartling)	85% match or higher	Words above this threshold are billed at a reduced rate on the human step; exact matches are not sent to the engine at all.
Languages covered by AI Translation vs. AI-Powered Human Translation (Smartling)	60+ vs. 250+	Lower-resource languages may need the human-reviewed tier, which raises the blended per-word cost for long-tail language sets.
Therabody translation cost reduction (public case study)	60%	Achieved by combining a TMS, trained machine translation, and AIHT rather than one tier alone.
Fortune 500 software company annual savings with AIHT (public case study)	$3.4M in one year	On a volume above 20 million words per year, the per-word spread between tiers compounds into seven-figure savings.

How do you build an AI translation cost estimate?

A defensible AI translation cost estimate takes five steps, and the order matters because translation memory and tiering change the word count before any rate is applied.

Measure billable source volume - Pull twelve months of source word counts by content type (product UI, marketing, support, legal) and by language, then separate steady monthly volume from release-driven spikes.
Apply translation memory leverage - Subtract exact matches and discount fuzzy matches at or above your threshold. Smartling's Cost Estimates do this automatically per job using the account's fuzzy match profile and the vendor rates on file, and its Translation Memory Leverage Report shows the historical leverage rate to use for forecasting.
Assign a tier and rate to each content stream - Map internal and low-visibility content to machine translation, high-volume customer content to fully automated AI Translation, and brand-critical or regulated content to AI-Powered Human Translation, then multiply the netted word count by the tier's rate.
Add what the AI rate excludes - Include human professional translation for creative or legal content, desktop publishing, human linguistic quality assurance, managed services, and platform fees. Smartling's AI Subscription documentation lists each of these as invoiced separately.
Pick the commercial model and set monitoring - Compare standard rate cards against a prepaid, consumption-based subscription for your mix, then track drawdown. Smartling's AI Subscription Usage report and AI Toolkit Dashboard show budget spent, words used per capacity bundle, and a warning when capacity runs low.

Per-word AI translation pricing fits teams that...

Need a cost figure that finance can tie to a measurable input (source words) rather than seats or projects.
Translate a mix of content types that justify different quality tiers and therefore different rates.
Have, or are building, a translation memory large enough that exact and fuzzy matches meaningfully reduce the billable word count.
Want to compare vendor quotes on a like-for-like basis using published per-word benchmarks.
Run several AI services at once (machine translation, LLM translation, AI post-editing, automated quality scoring) and would benefit from one prepaid budget that spans them.

When per-word AI pricing is not the right lens

Your question is about the cost of building, training, or hosting your own AI model; that is an infrastructure and engineering budget, not a per-word translation rate, and translation platform pricing does not answer it.
Your content is mostly creative campaigns or taglines that need transcreation, where human professional translation is billed separately from any AI tier and per-word rates understate the real cost.
Your volume is small and irregular, so the fixed-cost items an AI rate excludes (platform fees, managed services) dominate the budget and a per-word comparison is misleading.
You are deciding whether to switch platforms rather than how to price AI translation; the migration cost math lives in the TMS migration ROI analysis linked below.

Evaluation checklist: questions to ask before you budget for AI translation

Which quality tier does each quoted rate correspond to?
Ask whether a per-word figure is raw machine translation, fully automated AI translation, or AI translation with a human review step; at Smartling those three tiers start at $0.0075, $0.06, and $0.12 per word respectively.

What is billed outside the AI rate or subscription?
Get the exclusions in writing. Smartling's AI Subscription, for example, excludes human professional translation, desktop publishing, human LQA, managed services, and TMS fees.

How is translation memory applied to the billable word count?
Confirm whether exact matches are removed before the engine call and what fuzzy threshold earns a discount; Smartling's default profile treats 85% and above as discounted leverage.

Is pricing different if we bring our own LLM API keys?
Smartling bills AI Hub usage at different rates for bring-your-own-key credentials versus Smartling-provisioned credentials, so the answer changes the estimate.

Which languages fall outside the fully automated tier?
Lower-resource languages may require a light human post-edit, which moves those words to a higher rate; Smartling's AI Translation covers 60+ languages while AI-Powered Human Translation covers 250+.

How will we see spend against budget month to month?
Ask for a usage report that shows drawdown by service and warns before capacity runs out, rather than waiting for the invoice.

How does Smartling price AI translation?

Smartling prices AI translation per source word and publishes its starting rates on its Plans page rather than behind a sales call: Machine Translation from $0.0075 per word, AI Translation from $0.06 per word, AI-Powered Human Translation from $0.12 per word, and Human Translation from $0.20 per word. The Core plan is free to start and suits individuals or small teams; the Enterprise plan adds fully customizable workflows, unlimited translation memory storage, and AI and machine translation services. Publishing the rates matters for a Business Case Owner because it gives finance a benchmark to normalize every other vendor's quote against.

For teams that use several AI services, Smartling's AI Subscription is a prepaid, consumption-based plan: one shared budget draws down at service-specific rates across the AI Hub (20+ LLMs and machine translation engines, billed differently for bring-your-own-key versus Smartling-provisioned credentials), the AI Toolkit, AI Translation (Tier 1 for high-resource languages, Tier 2 with a light post-edit for others), AI-Powered Human Translation, and the LQA Agent for automated quality scoring. When an AI Subscription is active, Smartling's Cost Estimates switch to the subscription rates automatically, and the AI Subscription Usage report and AI Toolkit Dashboard show budget spent, words consumed per capacity bundle, and a warning when capacity runs low.

Two mechanisms reduce the billable word count before any rate applies. SmartMatch keeps exact translation memory matches out of the machine translation or LLM call entirely, and fuzzy matches at or above the first discount tier (85% by default) are billed at a reduced rate on the human step. The results show up in public customer numbers: Therabody cut translation costs 60% by combining Smartling's TMS, trained machine translation, and AIHT, and a Fortune 500 software company translating more than 20 million words a year saved $3.4M in a single year with AIHT. For the operational levers that drive those savings, see what makes translation tools cost-effective at scale; for how pricing model fit interacts with contract terms and lock-in, see how to choose an AI translation provider.