Smartcat vs Unbabel vs DeepL: Real AI Translation Cost per 1 Million Words
Short answer
These three products sit at three different points on the automation-to-human-oversight spectrum, and that difference matters more than any single price comparison. DeepL is pure machine translation, billed by the character, with a genuinely public rate card. Smartcat is a consumption-based platform billed in its own "Smartword" unit, with a published entry price but no fully published per-word rate beyond that entry point. Unbabel is a fully managed, human-in-the-loop service, entirely quote-based, with reported contract values ranging from $10,000 to $125,000 a year depending on volume and quality requirements.
Pricing at a glance
| Smartcat | Unbabel | DeepL | |
|---|---|---|---|
| Model | Consumption-based, billed in "Smartwords" (roughly one Smartword per source word), with unlimited users on every plan | Fully managed machine-translation-plus-human-post-editing service | Pure machine translation, billed per character via API, or per-user subscription via its Pro apps |
| Public rate card | Partial — a published entry point exists, but the per-Smartword rate above that isn't fully public | None — quote-only | Yes — confirmed API rates |
| Free tier | "Forever Free" plan with 15,000 words/month | Not confirmed | Developer tier: 1 million characters, one-time (not renewing monthly) |
| Entry price | Reported at $1,200/year for a Basic plan (Organizations or Agencies) | Reported median contract value: $30,000/year, per third-party transaction data | Growth: $26/month (annual billing), including roughly 12 million characters/year, up to a documented 50-million-character-per-month cap |
| Human-in-the-loop option | Yes — Marketplace access to 500,000+ vetted human linguists, copywriters, and reviewers, priced per the individual linguist's own rate | Yes — this is the core of Unbabel's model, not an add-on | No — DeepL is machine-translation-only; it does not offer a managed human-review service of its own |
| Reported range at scale | Not confirmed beyond the entry price | $10,000–125,000/year, per third-party benchmark data | Growth's documented ceiling is 50 million characters/month; above that, Enterprise pricing is custom-quoted |
Characters vs. words — why this comparison needs an explicit conversion
DeepL bills in characters. Smartcat bills in its own "Smartword" unit, which corresponds roughly to one source word. Unbabel's pricing is described as depending on volume typically measured in words or characters, without a single fixed unit. To compare all three against the same "per 1 million words" basis, this article uses a standard approximation of roughly 5.5–6 characters per word (including spaces) — a common rule of thumb for English-language content, though actual character-to-word ratios vary by language and content type. Every DeepL figure below that starts from a word-based scenario is therefore a converted estimate, not a native DeepL unit, and is labeled as such.
What headline pricing excludes
Smartcat's per-Smartword consumption rate beyond its published entry-level plan pricing is not clearly published. The platform's own documentation confirms that translation-memory matches, repeated segments, and auto-copy-from-source don't consume Smartwords on current plans — a genuinely useful cost-reduction mechanism, since effective cost per word falls as an organization's translation memory grows — but the actual dollar-per-Smartword rate for content that does consume Smartwords isn't stated in a single, clear published figure across current pricing materials.
Unbabel's pricing explicitly varies by quality tier and turnaround-time requirement, not just volume. A company assuming a linear per-word rate will find Unbabel's actual cost structure more variable than that — the same word volume can cost meaningfully different amounts depending on how much human post-editing and how fast a turnaround the content requires.
DeepL's Growth tier has a documented cap (50 million characters/month) beyond which Enterprise pricing takes over, and DeepL does not publish an exact overage rate for volume between its included annual allowance and that cap. A company modeling DeepL's cost at high volume should treat the Growth-tier headline price as a floor, not a complete number, once monthly usage exceeds the roughly 1-million-character-per-month average implied by the annual plan.
Hidden costs
- Smartcat's Marketplace human-expert pricing is set by individual linguists, not by Smartcat itself — a company relying on Marketplace for human review or translation should expect real per-project rate variation depending on which specific expert accepts the work, language pair, and subject-matter specialization.
- Unbabel's setup and integration costs are described as a real, separate cost category on top of the ongoing per-word-equivalent service cost, according to third-party benchmark analysis of the category.
- DeepL's document-translation feature carries its own per-document minimum character charge — a very short document is billed as if it were substantially longer, a real cost consideration for a workflow translating many short files rather than fewer long ones.
Worked scenarios
100,000 words/month
Converted to characters at roughly 5.5–6 characters/word, this is approximately 550,000–600,000 characters/month — comfortably within DeepL's Growth-tier average annual allowance (~1 million characters/month), so DeepL's cost here is the flat $26/month Growth price. Smartcat's published entry-level plan ($1,200/year, or $100/month) is priced well above this specific volume if the entry plan's included Smartword allowance covers it — though the exact allowance wasn't confirmed from available sources, so this figure should be treated as the plan's flat price rather than a confirmed per-word-justified cost at this volume. Unbabel, given its reported $10,000–125,000/year range skews toward larger enterprise deployments, is likely not economically positioned for this volume unless the specific use case (e.g., high-stakes customer support translation needing human review) specifically justifies its managed-service model regardless of raw word count.
1,000,000 words/month
Converted to characters, this is approximately 5.5–6 million characters/month — exceeding DeepL Growth's roughly 1-million-character-per-month average allowance, meaning this volume would incur real overage charges within a single month, at a rate DeepL does not publicly specify. Smartcat's cost at this volume depends on the unconfirmed per-Smartword rate beyond its entry tier; a company at this volume should request a specific quote rather than assume the $1,200/year entry price scales linearly. Unbabel's reported median contract value ($30,000/year) becomes a more plausible reference point at this volume, particularly if human post-editing or customer-facing quality requirements are part of the specification — though the exact word volume implied by that median contract wasn't confirmed from available sources.
5,000,000 words/month
Converted to characters, this is approximately 27.5–30 million characters/month — still within DeepL's documented 50-million-character-per-month Growth-tier cap, but representing substantial overage beyond the included annual allowance; DeepL Enterprise pricing may be the more appropriate and cost-effective path at this volume, though it is entirely custom-quoted. Smartcat, at this volume, is very likely to require its Enterprise/custom tier rather than the published Basic entry plan. Unbabel's reported range tops out at $125,000/year in third-party benchmark data — a company at this volume needing human-reviewed, managed translation should expect to be evaluated toward the higher end of that reported range, or beyond it, depending on quality tier and language-pair complexity.
Normalizing across billing units
The clearest structural distinction in this article isn't between prices — it's between what's being purchased. DeepL sells pure machine translation, metered precisely by character, with no human-review component of its own. Smartcat sells a platform with metered machine translation plus optional, separately-priced access to human experts. Unbabel sells a fully managed outcome — translated content at a specified quality bar — with the underlying mix of machine translation and human post-editing handled internally and not separately itemized to the customer. A "cost per word" comparison across all three is only fair once a buyer has decided which of these three purchasing models actually fits the need — pure MT, platform-plus-optional-humans, or fully managed service.
Break-even and crossover
The clearest calculable question in this category is not a crossover between these three vendors specifically, but a threshold: at what volume does DeepL's per-character overage cost (once the Growth-tier annual allowance is exceeded) become expensive enough that Smartcat's platform-plus-marketplace model, or Unbabel's fully managed service, become worth evaluating instead — even though this article cannot state that threshold precisely, since DeepL's overage rate isn't published and Smartcat's per-Smartword rate beyond entry-level isn't fully public either. This is a genuine finding about the category's opacity rather than a gap this article can responsibly fill with an invented number.
Who pays more, and when
- A team needing pure, fast, high-quality machine translation with no human review requirement is well served by DeepL, with genuinely transparent per-character pricing at low-to-moderate volume, and a real, if imprecisely priced, overage consideration at higher volume.
- A team needing a platform that combines machine translation with optional, on-demand human expert review fits Smartcat's model, though budgeting precisely requires a direct quote once volume exceeds the published entry tier's likely allowance.
- A team needing translated content delivered as a finished, quality-guaranteed outcome, without managing the machine-translation-versus-human-review tradeoff themselves, is the scenario Unbabel is built for — at a real cost premium reflected in its reported $10,000–125,000/year range, justified by the managed-outcome model rather than raw per-word economics.
- A team price-sensitive at high volume should specifically press DeepL and Smartcat for their exact overage and above-entry-tier per-unit rates before committing, since both remain genuinely unclear from public information at the volumes where they'd matter most.
Limitations and uncertainty
DeepL's API rates are confirmed from its own pricing documentation, though its exact overage rate on the Growth tier, once a customer exceeds the included annual allowance within a single month, is not published and is not estimated in this article. Smartcat's entry-level plan price ($1,200/year) is reported consistently across multiple sources, but its per-Smartword consumption rate beyond that entry tier is not clearly published in a single, current, citable figure. Unbabel does not publish pricing at all; the $10,000–125,000/year range and $30,000/year median figure used throughout this article are third-party transaction benchmark data, not official rates, and the specific word volumes implied by those contract values were not confirmed from available sources. All character-to-word conversions in this article use a standard approximation (5.5–6 characters per word) that varies by language and content type and should be treated as directional rather than exact.
Official sources
- deepl.com/pro-api (confirmed API pricing)
- smartcat.com/pricing, help.smartcat.com (confirmed entry-tier pricing and consumption mechanics; per-Smartword rate beyond entry tier not published)
- Unbabel does not publish pricing; figures in this article are drawn from third-party transaction benchmark data