DeepL vs Google Cloud Translation vs Azure AI Translator: Real Cost per Million Characters in 2026
Short answer
At low volume, Azure's free tier (2 million characters/month) covers most small workloads entirely — a genuinely underrated fact. At moderate-to-high volume, Azure's standard per-character rate ($10/1M) is also half of Google's ($20/1M), making Azure the cheapest of the three at every volume level examined in this article — not just at the low end. DeepL is the most expensive per character on its API, and it also just went through a real pricing-structure change: its old API Free/Pro plans were retired in July 2026, replaced by Developer and Growth tiers with different mechanics — if you're reading an older comparison, the numbers in it are stale.
Pricing at a glance
Rates are per million characters unless noted. All three offer a genuine free allowance before any billing starts.
| DeepL API | Google Cloud Translation | Azure AI Translator | |
|---|---|---|---|
| Free tier | Developer: 1M characters, one-time (not monthly) | 500,000 characters/month | 2,000,000 characters/month (F0 tier) |
| Standard translation rate | Growth: $26/month (annual billing), includes ~12M characters/year, up to 50M characters/month cap, then overage | $20/1M (Standard NMT) | $10/1M (Text Translation, S1 tier) |
| Higher-quality/LLM mode | Not offered as a separate tier | $10 input + $10 output per 1M (LLM Translation mode) | Not offered as a separate tier |
| Custom/fine-tuned model | Enterprise only, custom-quoted | Adaptive Translation: $25 + $25/1M; AutoML: $80/1M | Custom Translator inference: $40/1M |
| Document translation | Bundled into Pro subscriptions (separate from API); file-size caps by tier | Same per-character rate as text | $15/1M, plus $8/1,000 images for image-based documents |
| Custom model training | N/A | Included in AutoML/Adaptive pricing above | $10/1M source+target characters, capped at $300/training run |
| Custom model hosting | N/A | Not separately itemized in public pricing | $10/hosted model/region/month flat fee — charged even with zero usage |
A genuinely important currentness note on DeepL: as of July 2026, DeepL retired its old API Free (500K/month free) and API Pro ($5.49/month + $25/1M characters) plans. New API customers are directed to Developer (a one-time 1M-character allowance, not renewing monthly) and Growth ($26/month on annual billing, including roughly 12M characters/year, with a documented cap of 50M characters/month before Enterprise pricing applies). A lot of comparison content still circulating quotes the old API Pro rate — treat any DeepL API figure that doesn't mention Developer/Growth as outdated.
What headline pricing excludes
DeepL's Pro subscriptions and its API are two completely separate products with separate bills. A company using DeepL through its web app or browser extension (Starter/Advanced/Ultimate, roughly $10–70/user/month depending on tier) gets zero API character allowance from that subscription — a team needing both human-facing translation and programmatic API access pays for both separately.
Google's Standard NMT rate isn't the only way to get a translation from Google. The newer LLM Translation mode, priced at $10 input plus $10 output per million characters (effectively $20/1M round-trip, comparable to Standard), targets higher-nuance output — it's not automatically cheaper or more expensive than Standard NMT, it's a different quality tier at a similar price point.
Azure's free 2 million characters/month is real, but it's Microsoft's F0 tier specifically — it applies to text translation and shares its quota with custom-model training characters. A workload that both translates text and trains a custom model in the same month draws down the same free allowance for both.
Hidden costs
- Azure's Custom Translator model hosting fee is a flat $10/month per model per region, charged whether or not the model is used that month. A company that trains a custom model and then translates infrequently still pays this fee every month the model stays hosted.
- Azure's custom-model inference rate is roughly 4x its standard rate ($40/1M vs. $10/1M) — a custom-trained model isn't a free upgrade over the standard engine, it's a materially more expensive one, on top of the separate training and hosting fees.
- DeepL bills a minimum of 50,000 characters per translated document file, regardless of how short the actual document is — a one-page cover letter at DeepL's document-translation rate costs the same as a much longer document under that minimum.
- Google's Adaptive Translation and AutoML options both carry their own separate per-character rates ($25+$25/1M and $80/1M respectively) on top of Standard NMT — these aren't add-ons to the standard rate, they're distinct, more expensive billing paths for custom-data fine-tuning.
Worked scenarios
1 million characters/month
| Cost | |
|---|---|
| DeepL API (Growth, annual) | $26/month flat (this volume fits within the ~12M/year included allowance) |
| Google Cloud (Standard NMT) | 500K free + 500K × $20/1M = $10/month |
| Azure AI Translator | Fully covered by the 2M-character free tier = $0/month |
At this volume, Azure's free tier is the standout — a workload of 1 million characters a month costs nothing at all on Azure, while it costs $10 on Google and $26 on DeepL.
10 million characters/month
| Cost | |
|---|---|
| DeepL API (Growth, annual) | $26/month base, but this volume in a single month exceeds the ~1M/month average implied by the annual allowance — expect meaningful overage charges; DeepL does not publish an exact overage rate for Growth, so treat this as $26 plus an unconfirmed overage amount, not a final figure |
| Google Cloud (Standard NMT) | 500K free + 9.5M × $20/1M = $190/month |
| Azure AI Translator | 2M free + 8M × $10/1M = $80/month |
Azure is the clear winner at this volume among the two vendors with fully confirmed rates; DeepL's exact cost here can't be stated precisely without confirming its Growth-tier overage rate directly.
100 million characters/month
| Cost | |
|---|---|
| DeepL API | Exceeds Growth's documented 50M-character monthly cap — this volume requires DeepL's custom-quoted Enterprise tier; no public rate exists |
| Google Cloud (Standard NMT) | 500K free + 99.5M × $20/1M = $1,990/month |
| Azure AI Translator (pay-as-you-go) | 2M free + 98M × $10/1M = $980/month |
At high volume, Azure remains roughly half the cost of Google's standard rate, and DeepL drops out of the pay-as-you-go comparison entirely — anyone translating 100 million characters a month on DeepL is negotiating an Enterprise contract, not paying a published rate.
Normalizing across billing units
Cost per 100 translated documents is hard to state as a single number across vendors, because DeepL's document minimum (50,000 characters/document) makes its effective per-document cost highly dependent on actual document length — a batch of very short documents costs disproportionately more per document on DeepL than on Google or Azure, which bill purely by character count with no per-document floor.
Effective cost after fixed fees matters most for DeepL and Azure's custom-model path: DeepL's $26/month Growth base fee and Azure's $10/month-per-model hosting fee both mean the effective per-character rate is higher than the headline usage rate at low volume, and converges toward the headline rate only once volume is high enough to dwarf the fixed component.
Break-even and crossover
The clearest calculable crossover in this article: Azure's free 2M-character tier versus Google's 500K-character free tier. Any workload under 2 million characters/month costs $0 on Azure and something on Google (500K free, remainder at $20/1M) — Azure's larger free allowance is a straightforward, quantifiable advantage for any workload under that 2M-character line, not a marketing footnote.
Who pays more, and when
- Low-volume workloads (under 2M characters/month) are effectively free on Azure and cheap on Google; DeepL's $26/month Growth minimum makes it the most expensive of the three at this scale regardless of actual usage.
- Mid-to-high-volume, standard-quality translation favors Azure's $10/1M rate over Google's $20/1M — a straightforward 2x difference at the same quality tier.
- Workloads specifically needing DeepL's translation quality for European languages (a real, often-cited strength) have to accept its higher API cost and its 50M-character monthly ceiling as the price of that quality — there's no way to get DeepL's engine at Azure's or Google's rate.
- Custom, domain-fine-tuned translation is expensive on both Google (AutoML $80/1M) and Azure ($40/1M inference plus training and hosting fees) — this is a fundamentally pricier category than standard machine translation on any vendor, not a place to expect standard-tier pricing.
Limitations and uncertainty
DeepL's exact overage rate on the Growth tier, once a customer exceeds their included annual allowance within a single month, was not confirmed from an official source — treat the 10-million-character scenario above as a floor ($26) rather than a complete figure. DeepL's Enterprise pricing for volumes above the Growth tier's 50-million-character monthly cap is entirely custom-quoted and not estimated here. All Azure figures reflect standard pay-as-you-go (S1) rates; Azure also offers committed-use tiers at higher volumes (250 million characters/month and above) with lower effective per-character rates that weren't fully itemized here — a workload consistently above 100 million characters/month should get an Azure commitment-tier quote before assuming the pay-as-you-go rate applies.
Official sources
- deepl.com/pro-api (pricing), deepl.com/pro (subscription pricing)
- cloud.google.com/translate/pricing
- azure.microsoft.com/en-us/pricing/details/translator/, learn.microsoft.com (Azure AI Translator documentation)