# DeepL

> Translates text and documents between languages using its own models, and lets company terminology be fixed through glossaries and style rules.

- Vendor: DeepL SE, Köln
- Canonical URL: https://www.convios.com/en/toolbox/deepl
- Language version: https://www.convios.com/de/werkzeugkasten/deepl
- Area: Content & web · Cluster: translation
- Role: Off-the-shelf product · Origin: AI-native
- As of: 2026-08-06 · Reviewed: 2026-08-06 · Author: Dr. Oliver Gausmann, Convios GmbH
- Toolbox: https://www.convios.com/en/toolbox — Markdown: https://www.convios.com/en/toolbox.md

## Verdict

Efficiency only: Translation is a cost item, and a competitor signs up for the same subscription on the same day. What prevents falling behind is the ability to publish continuously in several languages and to check the quality while doing so. That ability is homework, and DeepL is one purchasable part of it. Where translation sits on the product surface, quality weighs more heavily, but the distance then comes from your own terminology and your own review loop.

## Suitability by company size

- Solo: suitable — The free version covers 50,000 characters a month. Anyone translating documents regularly ends up at 7.49 euros a month with 300,000 characters, and a single seat rarely needs more.
- Mid-market: suitable — 24.99 euros per user per month on the Team plan stays predictable. The friction sits in sign-in: single sign-on is only available on Team and Business above 50 users, and below that IT manages accounts by hand.
- Enterprise: suitable with caveats — Functionally the tool scales across the organisation. The hurdles sit in the contract: user provisioning via SCIM and domain management from Business upwards, customer-managed keys only on the Enterprise plan, and since the infrastructure change a fixed EU region as a separately purchased add-on for sales-assisted contracts.

## Vendor staying power

Established: DeepL Translator has been publicly available since 2017, putting the product over nine years in the market. The company is registered in Cologne as a European public limited company (Societas Europaea) and emerged from DeepL GmbH. Security attestations have been running since January 2021 and continue through 2026. The most recent publicly quantified funding dates from 22.05.2024. Group revenue and profit figures are not publicly retrievable, so the rating rests on time in market and the company register.

- Legal form and company register entry: DeepL SE, Maarweg 165, 50825 Cologne, entered at the Amtsgericht Köln (local court) under HRB 104617, VAT ID DE349242045. Dr. Jaroslaw Kutylowski is named as authorised representative and responsible for content. (source: https://www.deepl.com/de/publisher, as of 2026-08-06)
- Company register history: First registered on 03.02.2021 as Blitz K21-863 SE, trading as DeepL SE since 10.01.2022, formed by merger out of DeepL GmbH. The 2024 annual accounts were filed on 05.05.2026, the consolidated 2023 accounts on 06.05.2026. (source: https://www.northdata.de/?id=DER3306.HRB104617, as of 2026-08-06)
- Time in market: According to the company, DeepL Translator has been offered at deepl.com since 2017. The statement comes from a dated vendor press release of 19.01.2021. (source: https://www.deepl.com/en/press-release/deepl-achieves-iso-27001-certification, as of 2021-01-19)
- Most recent publicly quantified funding: USD 300 million, announced on 22.05.2024, at a post-money valuation of USD 2 billion. Led by Index Ventures, with ICONIQ Growth, Teachers' Venture Growth and existing investors IVP, Atomico and World Innovation Lab. A report carrying the same figure in November 2025 restates the same round. (source: https://www.ots.at/presseaussendung/OTS_20240522_OTS0186/deepl-verkuendet-investment-in-hoehe-von-300-millionen-us-dollar-bei-einer-gesamtbewertung-von-zwei-milliarden-us-dollar, as of 2024-05-22)
- Security attestations over time: ISO 27001 first reported on 19.01.2021, issued by TÜV Rheinland Cert after an audit in November 2020, covering development, operation and distribution of the translation software. An attestation under the BSI C5 Type 2 criteria catalogue is documented in a vendor post dated 03.10.2025. As of 06.08.2026 the trust portal lists ISO/IEC 27001:2022, SOC 2 Type 2 and C5. (source: https://www.deepl.com/en/press-release/deepl-achieves-iso-27001-certification, as of 2021-01-19)
- Infrastructure shift to a hyperscaler: Trust portal notice of 06.02.2026: Amazon Web Services is added as a sub-processor, effective for new contracts and renewals from 01.01.2026. Customer content is then processed across AWS regions worldwide by default. The company's own data centres in Iceland and Sweden remain in use for research, model training and free-tier usage. (source: https://deepl.safebase.us/, as of 2026-02-06)
- Customer count according to the vendor: On its pricing page the vendor cites more than 200,000 businesses worldwide as of 06.08.2026. The funding announcement of 22.05.2024 said more than 100,000 businesses, governments and other organisations were using DeepL. Both are vendor figures with no disclosed method of counting. (source: https://www.ots.at/presseaussendung/OTS_20240522_OTS0186/deepl-verkuendet-investment-in-hoehe-von-300-millionen-us-dollar-bei-einer-gesamtbewertung-von-zwei-milliarden-us-dollar, as of 2026-08-06)

## Cost of leaving

Moderate: The translations already sit in your own documents, so switching engines costs no data migration. What stays behind with the vendor is the accumulated stock: glossaries, style rules and the translation memory on the upper plans, plus the integrations into Word, Outlook and translation environments. Anyone switching re-checks quality against their own texts and rebuilds the terminology.

## Regulation and data

| Point | Finding | Evidence | As of |
|---|---|---|---|
| Data processing agreement | contractually provided for, concluded via sales — For the Article 28 GDPR agreement the privacy policy points to sales@deepl.com, and the Pro terms provide for it in section 8.1. No publicly retrievable contract document exists; templates sit behind the access request in the trust portal. | partially evidenced | 2026-08-06 |
| Storage location | AWS regions worldwide by default, fixed EU region via the Data Residency add-on — Until the change, services ran from the company's own data centres in Europe. Since 01.01.2026 new contracts and renewals fall under the new infrastructure terms, under which customer content is processed across several AWS regions by default. Pinning to one region is a paid add-on for sales-assisted Team, Business and API for Business contracts and is included on the Enterprise plan. Customers objecting to the new terms stay on the previous infrastructure until their contract ends. | partially evidenced | 2026-02-06 |
| Subprocessors | named individually, full list only after an access request — Amazon Web Services is named as a sub-processor, and for DeepL Voice for Meetings additionally Microsoft Ireland Operations Limited and Amazon Web Services EMEA SARL, with purpose and registered office. The only freely visible list in the trust portal covers Early Access Programs; the remaining documents require an access request. | partially evidenced | 2026-08-06 |
| Third-country transfer | transfers occur, legal basis named — For international transfers the vendor names the EU-US Data Privacy Framework and standard contractual clauses. Which recipient operates on which of the two bases is not publicly assigned. According to the vendor, the transfer impact assessment sits in the trust portal behind the access request. | partially evidenced | 2026-08-06 |
| Training on customer data | excluded on the paid plans, the default in the free version — Two statements on the same website part ways here. The security page says texts are never used to train the models without consent. In section 3 the privacy policy says that for the free services, uploaded content and its translations are processed for a limited period in order to train and improve the neural networks. For Translator Pro, API Pro and Write Pro, section 4 expressly rules out use for quality improvement, and the price list shows the exclusion from the Individual plan upwards. Under the terms of use the free version must not contain personal data in any case. | partially evidenced | 2026-08-06 |
| Retention and deletion | deletion after the service is rendered, saved translations 90 days after the contract ends — For the paid services the privacy policy commits to deleting all submitted texts and translations once the service has been fully rendered. The Pro terms quantify two further periods: content the customer deliberately saved is deleted 90 days after the contract ends, and in the event of error patterns the vendor may retain content encrypted for up to 72 hours. | evidenced | 2026-08-06 |
| Certifications | ISO/IEC 27001:2022, SOC 2 Type 2 and C5, evidence only after an access request — The trust portal lists the three attestations with standard and version and assigns them to the vendor itself. The certificates and audit reports sit behind an access request as private documents, so the current scope and audit period remain publicly open. What is publicly documented with a date is the first ISO 27001 attestation of 19.01.2021 by TÜV Rheinland Cert. | partially evidenced | 2026-08-06 |
| EU AI Act, Article 50 | vendor places itself under the regulation, commitment on Article 50 open — The vendor states that, being based in Germany, it is subject to the GDPR and the AI Act, and its trust portal carries a responsible AI statement. A named commitment on the transparency duties under Article 50 of Regulation (EU) 2026/1744 was not publicly findable. For translated text the labelling can be applied by the deploying organisation itself. | partially evidenced | 2026-08-06 |
| Audit logging | audit logs quantified on the API plans, open on the user plans — The API price list quantifies audit log retention per tier: 30 days on Developer, one year on Growth, two years on Enterprise API. For the user plans the price list names statistics and reporting from Team upwards without stating scope or retention period. User provisioning via SCIM and domain management are available from Business upwards. | partially evidenced | 2026-08-06 |

## Cost

- Entry: Free version with 50,000 characters a month. Individual costs 7.49 euros a month billed annually with 300,000 characters, Team 24.99 euros per user per month with one million characters, Business 49.99 euros per user per month with no character cap. On the API, Developer is free with one million characters and Growth costs 23.80 euros a month billed annually plus usage. All figures include VAT, as of 06.08.2026. (as of 2026-08-06)
- Where it gets expensive: What gets expensive is what procurement asks about. Single sign-on is available on Team and Business only above 50 users, and not at all below that. User provisioning via SCIM and domain management start at Business, customer-managed keys only at Enterprise. The Write Pro writing assistant costs 6.99 euros per user per month as an add-on. Pinning to a fixed region has cost extra since the infrastructure change and is only obtainable through sales.

## Three routes compared

### Translation environment with a language service provider

The established route is RWS Trados Studio or memoQ, run together with a language service provider. Large organisations stay with it because three things remain in house that a translation API does not supply: the translation memory as an owned asset, a contractually owed human review step, and a party you can hold liable for a wrong translation in a package insert or a contract. The price is a turnaround measured in days rather than seconds.

### A general-purpose language model over an API

Claude or GPT translate usably and additionally handle tone, shortening and audience. The price comparison only holds after measuring against your own texts, because DeepL bills per character and general models bill per token. Three things are given up: glossaries, style rules and translation memory as a finished product feature, the quantified deletion commitment from the Pro terms, and an attestation under the BSI C5 criteria catalogue, which German procurement asks about regularly.

### Your own translation pipeline in front of several models

Concretely: LiteLLM as a router in front of several models, Langfuse to evaluate the output, n8n or your own code for the pipeline, the glossary as your own table in the content system. Building it takes eight to twelve person-days, so at a daily rate of 1,000 euros between 8,000 and 12,000 euros, roughly equivalent to 33 Team seats for a year. What fails is the evaluation. Without a calibrated review loop, a drop in quality after a model change is first noticed by the customer. Add on-call duty for a pipeline that day-to-day work depends on, and maintenance after the person who built it has left.

Recommendation by size:

- Solo: Free version, and the Individual plan once you work with it regularly.
- Mid-market: Buy. Before ordering, check your user count against the threshold of 50 for single sign-on.
- Enterprise: Buy, but settle the data processing agreement and the region in writing before renewal, because the new infrastructure terms apply from 2026.

## Context

- Implements method: [Regulatory Density Test](https://www.convios.com/en/methods/regulatory-density-test) — The test asks how densely regulation sits around the use case, and that is exactly what decides the plan here: where personal or regulated content is translated, usability depends on the data processing agreement and the region, both obtainable only through sales.
- Implements method: [Calibrated Evaluation Loop for AI Outputs (Criteria Drift)](https://www.convios.com/en/methods/eval-calibration) — Translation quality cannot be compared without a calibrated review loop, and without that comparison the question of building your own pipeline in front of a general-purpose model cannot be answered.
- Implements method: [Demand Creation vs. Demand Capture](https://www.convios.com/en/methods/demand-creation-vs-capture) — Creating demand in a second language market means publishing there continuously, and the translation pipeline decides whether that becomes a routine or stays a project.
- Displaces: Manual translation for standard texts, Outsourced translation for internal communication

## Evidence

- Plans, character limits, thresholds for single sign-on and provisioning — https://www.deepl.com/de/pro (as of 2026-08-06)
- Handling of texts in free and paid use, data processing agreement, named sub-processors — https://www.deepl.com/de/privacy (as of 2026-08-06)
- Deletion periods, storage on customer request, retention in the event of error patterns — https://www.deepl.com/de/pro-license (as of 2026-08-06)
- Shift to AWS, default region behaviour, Data Residency add-on, list of certifications — https://deepl.safebase.us/ (as of 2026-02-06)
- Hybrid infrastructure, legal framework, bases for third-country transfers — https://support.deepl.com/hc/en-us/articles/26380849099932-DeepL-infrastructure-and-data-protection (as of 2026-08-06)
- Company name, register, representation — https://www.deepl.com/de/publisher (as of 2026-08-06)
- Audit logs per API tier, character allowances, price of the Growth tier — https://www.deepl.com/de/pro-api (as of 2026-08-06)