The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Dutch against 10 requirements in three levels: the solutions that meet them, what they cost and how fast they are, from the 2026 benchmark of 57 AI models and MT engines.
A solution is a model plus its harness: prompts, glossaries and other context, pre- and post-processing. It meets a quality level when it meets every requirement of the level.
| Requirement | Description | Met when |
|---|---|---|
| Basic requirements for any company | ||
| General translation quality | Translation errors of every type, such as mistranslations, omissions and grammar errors, classified by type and severity following MQM, the industry standard. | No critical error, at most 3 major errors in the four test texts (3,034 words) and at most 20 errors per 1,000 words, weighted by severity (minor 1, major 5, critical 10). |
| Standard requirements adds the rules companies set for their content | ||
| Terminology | The approved translation of every glossary term, in any of its correct forms. | 100% of the glossary terms. |
| Tone of voice | The formal register (u) in every sentence in the second person or the imperative. | 100% of those sentences. |
| Tags | Every HTML tag of the source, with its attributes, in the translation. | 100% of the tags. |
| Unicode character integrity | Every character intact, with no corrupted characters and no letter from another script that only looks like the right one. | 100% of the characters. |
| Locale conventions | The punctuation marks of the target language in place of the English ones. | No pass-or-fail bar. We report it as a share, because whether to localize punctuation is a matter of house style. |
| HTML entities | No HTML character code, such as the code for a quotation mark, that the source does not carry. | No added HTML entity. |
| Valid JSON | The translated JSON document still parses. | The document parses. |
| Spacing at tag boundaries | The spaces next to HTML tags stay as in the source. | 100% of the tag boundaries. |
| Advanced requirements adds a requirement for some companies and use cases | ||
| Gender neutrality | A gender-neutral translation wherever the English source leaves the gender of a person open, as in “your manager” or “the expert”. | 100% of the sentences that refer to a person. |
Pick a quality level, or your own requirements, bars and constraints, to see the shortlisted solutions that meet them in Dutch, with a column for each requirement you select. Under Optimize for, choose which of them comes first: the one with the fewest total errors, the lowest cost or the shortest time. Click a column heading to sort by it.
The shortlisted solutions on 13 models, with their Intento LQA results. Total errors per 1,000 words add the weighted general translation quality errors (minor 1, major 5, critical 10) to the missed glossary terms, sentences in the wrong register, tag errors and sentences that fail gender neutrality, each counted once. Formatting counts as met or missed and does not add to the total errors. The models that did not make the shortlist have automatic scores only: the shortlist. Gender neutrality: the Advanced requirements count the two best solutions of the language, with the gender-neutrality agent where they need it. The results of the other solutions are shown for information.
The two-step solution (translation + post-edit) on OpenAI GPT-5.5 and the one-step solution on OpenAI GPT-5.6 Sol meet every Standard requirement in Dutch, with 14.5 and 15.8 weighted errors per 1,000 words. They have 14.8 and 16.5 total errors per 1,000 words. The one on OpenAI GPT-5.6 Sol costs $6.44 per million characters against $9.25 and takes 58 seconds per 1,000 words against 65. So Optimize for selects the two-step solution for quality and the one-step solution for cost and latency.
Constraints leave few options. Under the Standard requirements, a cost cap of at most $6.43 per million characters or a time limit below 58 seconds per 1,000 words removes both solutions. A limit of one step leaves only the solution on OpenAI GPT-5.6 Sol. The three real-time (<2 sec) solutions on the shortlist, DeepL Next Gen, Amazon Translate and OpenAI gpt-oss-120b, all miss general translation quality, terminology and tag handling.
With the gender-neutrality agent, the two-step solution on OpenAI GPT-5.5 meets every Advanced requirement. The agent raises its cost from $9.25 to $9.66 per million characters. Its total errors rise from 14.8 to 15.8 per 1,000 words, and its weighted errors from 14.5 to 15.8 per 1,000 words, with three major errors against one.
Under the Advanced requirements, fewer errors come at a much higher cost. The three-step solution (translation + LQA + repair) on OpenAI GPT-5.6 Sol with the gender-neutrality agent has 2.6 total errors per 1,000 words, against 15.8 for the two-step solution on OpenAI GPT-5.5 with the agent. It costs $981 per million characters against $9.66, and takes 155 seconds per 1,000 words against 56. Optimize for selects it for quality, and the two-step solution for cost and latency.
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
two-step
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
three-step with the gender-neutrality agent
OpenAI GPT-5.5
two-step with the gender-neutrality agent
OpenAI GPT-5.5
two-step with the gender-neutrality agent
Optimize for selects one of the solutions that meet the level: the one with the fewest total errors, the lowest cost or the shortest time. Where no solution meets the level, the card shows the available solution with the fewest total errors and what it misses. Total errors count every requirement with countable errors.
| Quality level | Tier | Solutions | Simplest solution | Model | Total errors per 1,000 words | Cost per million characters | Seconds per 1,000 words |
|---|---|---|---|---|---|---|---|
| Basic | Tier 1: simple solution + backup | ★★ | one-step | OpenAI GPT-5.6 Sol | 16.5 | $6.44 | 58 |
| Standard | Tier 1: simple solution + backup | ★ ☆ | one-step | OpenAI GPT-5.6 Sol | 16.5 | $6.44 | 58 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | with the gender-neutrality agent | OpenAI GPT-5.5 | 15.8 | $9.66 | 56 |
One solution per level: the simplest that meets the level, the cheapest of them where several are equally simple. ★ a one-step solution, ☆ a solution with more steps or the gender-neutrality agent.
Under the Basic requirements, Optimize for selects OpenAI GPT-5.5 for quality and OpenAI GPT-5.6 Sol for cost and latency. From Basic to Standard, the one-step solution on OpenAI GPT-5.5 drops out on tag handling and formatting. The two-step solution on the same model meets both. From Standard to Advanced, gender neutrality removes both. Only the two-step solution on OpenAI GPT-5.5 and the three-step solution (translation + LQA + repair) on OpenAI GPT-5.6 Sol, both with the gender-neutrality agent, meet every Advanced requirement.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Mistral Large 3 | 0.74 | 28.7 | Misses a Standard requirement |
| xAI Grok 4.20 (reasoning) | 1.03 | 35.3 | Misses a Standard requirement |
| MoonshotAI Kimi K2.5 | 1.2 | 28.3 | Misses a Standard requirement |
| z.ai GLM-5 | 1.78 | 28.7 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.62 | 28.3 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.21 | 19.1 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 5.75 | 22.4 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 6.44 | 15.8 | Meets every Standard requirement |
| OpenAI GPT-5.5 | 9.25 | 14.5 | Misses a Standard requirement |
| Amazon Translate | 15 | 69.5 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 20.4 | Misses a Standard requirement |
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 28.3 |
| Tag handling | 0.7 |
| Gender neutrality | 0.7 |
| Tone of voice | 0.4 |
| Terminology | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Several one-step solutions come close on tag handling and terminology. Translations from OpenAI GPT-5.5 and xAI Grok 4.20 (reasoning) keep 99.6% of the tags, and from Anthropic Claude Sonnet 5, OpenAI GPT-4.1 and OpenAI gpt-oss-120b 99.2%. Translations from DeepL Next Gen and Cohere Command A Translate get 98.1% and 97.6% of the glossary terms right. The two-step solution (translation + post-edit) on OpenAI GPT-5.5 keeps 100% of the tags and meets formatting.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Dutch | Class |
|---|---|---|
| Amazon Translate | 69.5 | MT engines |
| xAI Grok 4.20 (reasoning) | 35.3 | LLMs |
| OpenAI gpt-oss-120b | 33.9 | Open-weight LLMs |
| Cohere Command A Translate | 32 | LLMs |
| z.ai GLM-5 | 28.7 | LLMs |
| Mistral Large 3 | 28.7 | LLMs |
| MoonshotAI Kimi K2.5 | 28.3 | LLMs |
| OpenAI GPT-4.1 | 28.3 | LLMs |
| Anthropic Claude Sonnet 5 | 22.4 | LLMs |
| DeepL Next Gen | 20.4 | LLM-based MT engines |
| Google Gemini 3.1 Pro Preview | 19.1 | LLMs |
| OpenAI GPT-5.6 Sol | 15.8 | LLMs |
| OpenAI GPT-5.5 | 14.5 | LLMs |
No translation into Dutch: SpeakLeash Bielik-11B-v3.0-Instruct (wrong language), Widn AI Tower Vesuvius (cut off part-way).
The 11 LLM solutions on the shortlist, one of them open-weight, have 14.5 to 35.3 weighted errors per 1,000 words, and two of them meet general translation quality. Translations from DeepL Next Gen, an LLM-based MT engine, have 20.4 weighted errors per 1,000 words, and from Amazon Translate, an MT engine, 69.5, with four critical errors. None of the other one-step solutions have a critical error.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Widn AI Tower Sugarloaf | 0.786 | 9 | 1.19 | 11 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.785 | 11 | 1.21 | 15 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.785 | 16 | 1.23 | 18 | No | Not shortlisted |
| Google Translation LLM | 0.784 | 17 | 1.55 | 33 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.783 | 18 | 1.26 | 22 | No | Not shortlisted |
| Google Gemma 4 | 0.783 | 19 | 1.19 | 10 | No | Not shortlisted |
| Microsoft Translator | 0.782 | 20 | 1.24 | 20 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.782 | 21 | 1.27 | 23 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.781 | 22 | 1.27 | 24 | No | Not shortlisted |
| Translated Lara 2 | 0.781 | 23 | 1.33 | 26 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.780 | 24 | 1.30 | 25 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.780 | 25 | 1.24 | 19 | No | Not shortlisted |
| Google Gemini 3.5 Flash | 0.779 | 26 | 1.37 | 31 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.777 | 27 | 1.24 | 21 | No | Not shortlisted |
| DeepSeek V3.2 | 0.777 | 28 | 1.18 | 9 | No | Not shortlisted |
| Supertext LLM | 0.775 | 29 | 1.35 | 29 | No | Not shortlisted |
| MiniMax M2.5 | 0.774 | 30 | 1.36 | 30 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.774 | 31 | 1.37 | 32 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.773 | 32 | 1.35 | 28 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.771 | 33 | 1.35 | 27 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.763 | 34 | 1.70 | 35 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.753 | 35 | 2.01 | 39 | No | Not shortlisted |
| Baidu Translate API | 0.749 | 36 | 1.71 | 36 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.749 | 37 | 1.78 | 37 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.740 | 38 | 1.64 | 34 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.733 | 39 | 2.01 | 38 | No | Not shortlisted |
| AppTek NMT | 0.675 | 40 | 3.64 | 41 | No | Not shortlisted |
| Cohere Tiny Aya | 0.673 | 41 | 3.41 | 40 | No | Not shortlisted |
| Widn AI Tower Anthill | 0.430 | 42 | 11.56 | 42 | No | Not shortlisted |
| DeepL Next Gen | 0.794 | 1 | 1.00 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.792 | 2 | 1.08 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | 0.792 | 3 | 1.04 | 2 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | 0.791 | 4 | 1.20 | 13 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-4.1 | 0.791 | 5 | 1.17 | 8 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.789 | 6 | 1.20 | 12 | Yes | Top 8 by COMET-Kiwi |
| z.ai GLM-5 | 0.787 | 7 | 1.20 | 14 | Yes | Top 8 by COMET-Kiwi |
| MoonshotAI Kimi K2.5 | 0.786 | 8 | 1.23 | 17 | Yes | Top 8 by COMET-Kiwi |
| Cohere Command A Translate | 0.786 | 10 | 1.07 | 3 | Yes | Top 8 by MetricX |
| OpenAI gpt-oss-120b | 0.785 | 12 | 1.22 | 16 | Yes | Real-time place |
| Mistral Large 3 | 0.785 | 13 | 1.12 | 5 | Yes | Top 8 by MetricX |
| OpenAI GPT-5.5 | 0.785 | 14 | 1.17 | 7 | Yes | Top 8 by MetricX |
| Amazon Translate | 0.785 | 15 | 1.16 | 6 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Adjusted NMT | 1 | 0.794 | 1 | 1.00 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 2 | 0.792 | 4 | 1.08 | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 3 | 0.792 | 2 | 1.04 | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 4 | 0.791 | 13 | 1.20 | Top 8 by COMET-Kiwi |
| OpenAI GPT-4.1 | Adjusted LLM | 5 | 0.791 | 8 | 1.17 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 6 | 0.789 | 12 | 1.20 | Top 8 by COMET-Kiwi |
| z.ai GLM-5 | Adjusted LLM | 7 | 0.787 | 14 | 1.20 | Top 8 by COMET-Kiwi |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 8 | 0.786 | 17 | 1.23 | Top 8 by COMET-Kiwi |
| Widn AI Tower Sugarloaf | Adjusted NMT | 9 | 0.786 | 11 | 1.19 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 10 | 0.786 | 3 | 1.07 | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 11 | 0.785 | 15 | 1.21 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 12 | 0.785 | 16 | 1.22 | Real-time place |
| Mistral Large 3 | Adjusted LLM | 13 | 0.785 | 5 | 1.12 | Top 8 by MetricX |
| OpenAI GPT-5.5 | Adjusted LLM | 14 | 0.785 | 7 | 1.17 | Top 8 by MetricX |
| Amazon Translate | Off-the-shelf NMT | 15 | 0.785 | 6 | 1.16 | Top 8 by MetricX |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 16 | 0.785 | 18 | 1.23 | Not shortlisted |
| Google Translation LLM | Off-the-shelf NMT | 17 | 0.784 | 33 | 1.55 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 18 | 0.783 | 22 | 1.26 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 19 | 0.783 | 10 | 1.19 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 20 | 0.782 | 20 | 1.24 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 21 | 0.782 | 23 | 1.27 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 22 | 0.781 | 24 | 1.27 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 23 | 0.781 | 26 | 1.33 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 24 | 0.780 | 25 | 1.30 | Not shortlisted |
| Tilde TildeOpen LLM engine | Adjusted NMT | 25 | 0.780 | 19 | 1.24 | Not shortlisted |
| Google Gemini 3.5 Flash | Adjusted LLM | 26 | 0.779 | 31 | 1.37 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 27 | 0.777 | 21 | 1.24 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 28 | 0.777 | 9 | 1.18 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 29 | 0.775 | 29 | 1.35 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 30 | 0.774 | 30 | 1.36 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 31 | 0.774 | 32 | 1.37 | Not shortlisted |
| SYSTRAN PNMT API | Adjusted NMT | 32 | 0.773 | 28 | 1.35 | Not shortlisted |
| ModernMT Enterprise Edition | Adjusted NMT | 33 | 0.771 | 27 | 1.35 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 34 | 0.763 | 35 | 1.70 | Not shortlisted |
| Google Cloud Advanced Translation API | Adjusted NMT | 35 | 0.753 | 39 | 2.01 | Not shortlisted |
| Baidu Translate API | Off-the-shelf NMT | 36 | 0.749 | 36 | 1.71 | Not shortlisted |
| Yandex Cloud Translate v2 | Adjusted NMT | 37 | 0.749 | 37 | 1.78 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 38 | 0.740 | 34 | 1.64 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 39 | 0.733 | 38 | 2.01 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 40 | 0.675 | 41 | 3.64 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 41 | 0.673 | 40 | 3.41 | Not shortlisted |
| Widn AI Tower Anthill | Adjusted NMT | 42 | 0.430 | 42 | 11.56 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.785 | — | 1.14 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.785 | — | 1.13 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.781 | — | 1.33 | Same model family as xAI Grok 4.20 (reasoning) |
Every model we got a translation from, as the shortlist compared it: off the shelf, without context in the request, or adjusted, with it. A model that did not make the shortlist is shown adjusted, unless off the shelf it scores higher by 0.03 or more on COMET-Kiwi. Ranks among 42 models, with each model family counted once, and the top 8 of each metric in bold. Higher is better on COMET-Kiwi, lower on MetricX. The scores rank models within this language and are not a requirement result.
The metrics and Intento LQA order the shortlist differently. The solution on DeepL Next Gen, first on both metrics, meets tone of voice but misses general translation quality, terminology, tag handling and formatting. Amazon Translate is on the list through MetricX, 6th there and 15th on COMET-Kiwi, and translations from it have 69.5 weighted errors per 1,000 words. The models nearest the cut are Widn AI Tower Sugarloaf, 9th on COMET-Kiwi, and DeepSeek V3.2, 9th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | $6.44 |
| Standard | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | $6.44 |
| Advanced | OpenAI GPT-5.5, two-step with the gender-neutrality agent | $9.66 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Mistral Large 3 | $0.74 | LLMs |
| xAI Grok 4.20 (reasoning) | $1.03 | LLMs |
| MoonshotAI Kimi K2.5 | $1.2 | LLMs |
| z.ai GLM-5 | $1.78 | LLMs |
| OpenAI GPT-4.1 | $2.62 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.21 | LLMs |
| Anthropic Claude Sonnet 5 | $5.75 | LLMs |
| OpenAI GPT-5.6 Sol | $6.44 | LLMs |
| OpenAI GPT-5.5 | $9.25 | LLMs |
| Amazon Translate | $15 | MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Fable 5 | 0.451 | LLMs |
| Anthropic Claude Opus 4.8 | 0.451 | LLMs |
| Anthropic Claude Sonnet 5 | 0.448 | LLMs |
| Mistral Medium 3.1 | 0.349 | LLMs |
| Mistral Large 3 | 0.346 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.328 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.328 | LLMs |
| MoonshotAI Kimi K2.5 | 0.313 | LLMs |
| AI21 Jamba Large 1.7 | 0.306 | LLMs |
| Widn AI Tower Sugarloaf | 0.306 | LLM-based MT engines |
| Alibaba Qwen3-235B-A22B | 0.305 | Open-weight LLMs |
| Widn AI Tower Anthill | 0.304 | LLM-based MT engines |
| DeepSeek V3.2 | 0.292 | LLMs |
| z.ai GLM-5 | 0.287 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.274 | LLMs |
| Google Gemini 3.5 Flash | 0.274 | LLMs |
| Google Gemma 4 | 0.272 | Open-weight LLMs |
| Cohere Command A Translate | 0.27 | LLMs |
| MiniMax M2.5 | 0.268 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.251 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.249 | LLMs |
| Cohere Tiny Aya | 0.248 | LLMs |
| OpenAI GPT-5.6 Terra | 0.245 | LLMs |
| OpenAI GPT-5.5 | 0.244 | LLMs |
| OpenAI GPT-4.1 | 0.243 | LLMs |
| OpenAI GPT-5.6 Sol | 0.243 | LLMs |
| OpenAI GPT-5.6 Luna | 0.242 | LLMs |
The tokenizers produce a median 0.29 tokens per character of Dutch, 1.2 times their count for English, and the tokenizer of Anthropic Claude Fable 5 produces 1.9 times as many tokens per character as that of OpenAI GPT-5.6 Luna. From Basic to Standard the cheapest stays the one-step solution on OpenAI GPT-5.6 Sol, at $6.44 per million characters. Under the Advanced requirements it is the two-step solution (translation + post-edit) on OpenAI GPT-5.5 with the gender-neutrality agent, at $9.66 per million characters.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | 58 |
| Standard | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | 58 |
| Advanced | OpenAI GPT-5.5, two-step with the gender-neutrality agent | 56 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| Google Translation LLM | 0.5 | LLM-based MT engines |
| Pangeanic PangeaMT | 0.5 | MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| ModernMT Enterprise Edition | 0.9 | MT engines |
| Amazon Translate | 1 | MT engines |
| Translated Lara 2 | 1.2 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 1.4 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 2.9 | MT engines |
| Oracle Cloud Language Service | 4.2 | MT engines |
| Tilde TildeOpen LLM engine | 7.9 | LLM-based MT engines |
| Widn AI Tower Anthill | 8 | LLM-based MT engines |
| SYSTRAN PNMT API | 8.8 | MT engines |
| Baidu Translate API | 12 | MT engines |
| Cohere Tiny Aya | 13 | LLMs |
| OpenAI GPT-5.6 Luna | 13 | LLMs |
| Google Gemini 3.5 Flash | 15 | LLMs |
| Supertext LLM | 15 | LLM-based MT engines |
| xAI Grok 4.20 (non-reasoning) | 15 | LLMs |
| Alibaba Machine Translation API | 20 | MT engines |
| OpenAI GPT-5.6 Terra | 25 | LLMs |
| OpenAI GPT-4.1 | 26 | LLMs |
| Anthropic Claude Haiku 4.5 | 27 | LLMs |
| Mistral Medium 3.1 | 34 | LLMs |
| Cohere Command A Translate | 37 | LLMs |
| MiniMax M2.5 | 38 | LLMs |
| Anthropic Claude Sonnet 5 | 42 | LLMs |
| Swiss AI Apertus-8B-2509 | 42 | Open-weight LLMs |
| AI21 Jamba Large 1.7 | 43 | LLMs |
| DeepSeek V3.2 | 43 | LLMs |
| Mistral Large 3 | 47 | LLMs |
| Widn AI Tower Sugarloaf | 49 | LLM-based MT engines |
| xAI Grok 4.20 (reasoning) | 51 | LLMs |
| Anthropic Claude Fable 5 | 51 | LLMs |
| Anthropic Claude Sonnet 4.6 | 55 | LLMs |
| Anthropic Claude Opus 4.8 | 57 | LLMs |
| OpenAI GPT-5.6 Sol | 58 | LLMs |
| OpenAI GPT-5.5 | 65 | LLMs |
| AppTek NMT | 71 | MT engines |
| Google Gemini 3.1 Pro Preview | 92 | LLMs |
| Alibaba Qwen3-235B-A22B | 122 | Open-weight LLMs |
| Google Gemma 4 | 183 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 195 | LLMs |
| z.ai GLM-5 | 220 | LLMs |
Real-time models are those the shortlist counts as real-time, under 2 seconds per text across the languages we timed. The chart shows the time of each model into Dutch.
Nine of the ten real-time models are MT engines or LLM-based MT engines, and the other is the open-weight OpenAI gpt-oss-120b. Translations from the three on the shortlist have 20.4 to 69.5 weighted errors per 1,000 words. Under the Basic and Standard requirements, Optimize for selects the one-step solution on OpenAI GPT-5.6 Sol for latency, at 58 seconds per 1,000 words. Under the Advanced requirements, it selects the two-step solution on OpenAI GPT-5.5 with the gender-neutrality agent, at 56 seconds per 1,000 words.
| The two best solutions | Solution | Sentences about people that are gender-neutral | Meets every Advanced requirement | Cost per million characters | Seconds per 1,000 words |
|---|---|---|---|---|---|
| OpenAI GPT-5.5 | two-step, with the gender-neutrality agent | 100% | ✓ | $9.66 | 56 |
| OpenAI GPT-5.6 Sol | three-step, with the gender-neutrality agent | 100% | ✓ | $981 | 155 |
The tone of voice we check in Dutch is the formal register, with u. Translations from Cohere Command A Translate have 97.9% of the sentences in the formal register, and from Amazon Translate 72.3%. For gender neutrality we check neutral nouns and rephrasing, with no default gender. The one-step solutions have a median 0.7 gender-neutrality errors per 1,000 words. With the gender-neutrality agent, 100% of the sentences about people in translations from both of the two best solutions are gender-neutral.
The three LQA judges agree on a median of 96% of the sentences of the 13 shortlisted solutions. With the automatic metrics in agreement too, the errors, the requirement results and the shortlist on this page rest on measurement we can rely on.
Leave feedback