The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into European Spanish 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 (usted) 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, such as the Spanish ¿ and ¡. | 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 European Spanish, 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 11 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.
Two one-step solutions meet every Standard requirement in European Spanish at almost the same cost, $3.74 per million characters for OpenAI GPT-5.6 Terra and $3.87 for Google Gemini 3.1 Pro Preview. Translations from OpenAI GPT-5.6 Terra take 17 seconds per 1,000 words against 73 and have 20.1 total errors per 1,000 words against 24.6, so Optimize for selects OpenAI GPT-5.6 Terra for quality, cost and latency alike.
Constraints narrow this quickly. Under the Standard requirements, a cost cap below $3.74 per million characters or a time limit below 17 seconds per 1,000 words removes both solutions, and the three real-time solutions on the shortlist, Microsoft Translator, Amazon Translate and DeepL Next Gen, all miss general translation quality and tag handling.
For the Advanced requirements, the gender-neutrality agent on OpenAI GPT-5.6 Terra raises the cost from $3.74 to $3.92 per million characters and the time from 17 to 24 seconds per 1,000 words, and its total errors fall from 20.1 to 6.3 per 1,000 words.
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
xAI Grok 4.20 (reasoning)
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step with the gender-neutrality agent
OpenAI GPT-5.6 Terra
one-step with the gender-neutrality agent
OpenAI GPT-5.6 Terra
one-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 | xAI Grok 4.20 (reasoning) | 26.6 | $1.01 | 61 |
| Standard | Tier 1: simple solution + backup | ★★ | one-step | OpenAI GPT-5.6 Terra | 20.1 | $3.74 | 17 |
| Advanced | Tier 3: one solution | ☆ | with the gender-neutrality agent | OpenAI GPT-5.6 Terra | 6.3 | $3.92 | 24 |
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.
Each level up removes one of the three solutions that meet the Basic requirements: tag handling the one on xAI Grok 4.20 (reasoning), with 99.2% of the tags kept, and gender neutrality the one on Google Gemini 3.1 Pro Preview, with 99.1% of the sentences about people gender-neutral. Under the Basic requirements, Optimize for selects xAI Grok 4.20 (reasoning) for cost, at $1.01 per million characters, and OpenAI GPT-5.6 Terra for quality and latency. Under the other two levels, it selects OpenAI GPT-5.6 Terra for all three.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| xAI Grok 4.20 (reasoning) | 1.01 | 15.2 | Misses a Standard requirement |
| AI21 Jamba Large 1.7 | 3.04 | 23.7 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 3.74 | 9.9 | Meets every Standard requirement |
| Google Gemini 3.1 Pro Preview | 3.87 | 14.8 | Meets every Standard requirement |
| Widn AI Tower Sugarloaf | 10 | 35.3 | Misses a Standard requirement |
| Microsoft Translator | 10 | 48.1 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 14.06 | 14.2 | Misses a Standard requirement |
| Amazon Translate | 15 | 61.3 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 19.1 | Misses a Standard requirement |
Among the one-step solutions, a higher price does not come with fewer errors. Anthropic Claude Opus 4.8 costs $14.06 per million characters and has 14.2 weighted errors per 1,000 words, Amazon Translate $15 and 61.3, and DeepL Next Gen $27.50 and 19.1. The four cheapest, all LLMs from $1.01 to $3.87, span 9.9 to 23.7 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 19.1 |
| Gender neutrality | 9.8 |
| Tag handling | 2.3 |
| Tone of voice | 2.1 |
| Terminology | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Terminology and tags are missed by small margins: Anthropic Claude Opus 4.8 and Cohere Command A Translate get 98.7% and 98.8% of the glossary terms right, and xAI Grok 4.20 (reasoning), AI21 Jamba Large 1.7 and DeepL Next Gen keep 99.2%, 98.6% and 96.5% of the tags. Tone of voice separates the MT engines: Microsoft Translator has 4.3% of its sentences in the tone of voice, Amazon Translate 74.5% and Elia Machine Translation System 76.6%, while every LLM solution reaches 100%.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in European Spanish | Class |
|---|---|---|
| Elia Machine Translation System | 72.5 | MT engines |
| Amazon Translate | 61.3 | MT engines |
| Microsoft Translator | 48.1 | MT engines |
| Widn AI Tower Sugarloaf | 35.3 | LLM-based MT engines |
| AI21 Jamba Large 1.7 | 23.7 | LLMs |
| Cohere Command A Translate | 19.1 | LLMs |
| DeepL Next Gen | 19.1 | LLM-based MT engines |
| xAI Grok 4.20 (reasoning) | 15.2 | LLMs |
| Google Gemini 3.1 Pro Preview | 14.8 | LLMs |
| Anthropic Claude Opus 4.8 | 14.2 | LLMs |
| OpenAI GPT-5.6 Terra | 9.9 | LLMs |
No translation into European Spanish: Widn AI Tower Vesuvius (cut off part-way).
The Intento LQA results group by model type in European Spanish: the six LLM solutions on the shortlist have 9.9 to 23.7 weighted errors per 1,000 words, the two LLM-based MT engines 19.1 and 35.3, and the three MT engines 48.1 to 72.5. All six shortlisted LLMs support the language by our measured criteria, and three of them meet general translation quality.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Google Gemini 3.5 Flash | 0.807 | 9 | 1.33 | 14 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.807 | 10 | 1.35 | 20 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.807 | 11 | 1.41 | 29 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.807 | 12 | 1.32 | 13 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.807 | 13 | 1.41 | 27 | No | Not shortlisted |
| OpenAI GPT-5.5 | 0.807 | 14 | 1.31 | 12 | No | Not shortlisted |
| Mistral Large 3 | 0.806 | 15 | 1.34 | 16 | No | Not shortlisted |
| DeepSeek V3.2 | 0.806 | 16 | 1.36 | 22 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.806 | 17 | 1.30 | 10 | No | Not shortlisted |
| z.ai GLM-5 | 0.805 | 18 | 1.36 | 21 | No | Not shortlisted |
| Meta Llama 4 Maverick | 0.804 | 19 | 1.41 | 28 | No | Not shortlisted |
| Google Gemma 4 | 0.803 | 22 | 1.35 | 19 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.803 | 23 | 1.30 | 9 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.802 | 24 | 1.37 | 25 | No | Not shortlisted |
| MiniMax M2.5 | 0.801 | 25 | 1.46 | 33 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.801 | 27 | 1.36 | 23 | No | Not shortlisted |
| Translated Lara 2 | 0.800 | 28 | 1.67 | 38 | No | Not shortlisted |
| Supertext LLM | 0.800 | 29 | 1.43 | 31 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.799 | 30 | 1.36 | 24 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.796 | 31 | 1.40 | 26 | No | Not shortlisted |
| Google Translation LLM | 0.796 | 32 | 1.74 | 40 | No | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | 0.795 | 33 | 1.43 | 32 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.793 | 34 | 1.42 | 30 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.792 | 35 | 1.49 | 35 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.791 | 36 | 1.35 | 18 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.790 | 37 | 1.47 | 34 | No | Not shortlisted |
| NiuTrans | 0.789 | 38 | 1.60 | 37 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.786 | 39 | 1.56 | 36 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.774 | 40 | 1.68 | 39 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.771 | 41 | 2.36 | 45 | No | Not shortlisted |
| iFlyTek MT | 0.771 | 42 | 1.87 | 41 | No | Not shortlisted |
| Tencent Cloud TMT | 0.765 | 43 | 1.98 | 42 | No | Not shortlisted |
| Naver Papago NMT Commercial | 0.764 | 44 | 2.03 | 43 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.759 | 45 | 2.05 | 44 | No | Not shortlisted |
| Cohere Tiny Aya | 0.751 | 46 | 2.38 | 46 | No | Not shortlisted |
| AppTek NMT | 0.699 | 47 | 3.32 | 47 | No | Not shortlisted |
| Widn AI Tower Anthill | 0.485 | 48 | 10.14 | 48 | No | Not shortlisted |
| DeepL Next Gen | 0.818 | 1 | 1.14 | 1 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | 0.811 | 2 | 1.30 | 11 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Terra | 0.811 | 3 | 1.29 | 8 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.810 | 4 | 1.27 | 4 | Yes | Top 8 by both metrics |
| Cohere Command A Translate | 0.810 | 5 | 1.33 | 15 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | 0.809 | 6 | 1.34 | 17 | Yes | Top 8 by COMET-Kiwi |
| Amazon Translate | 0.809 | 7 | 1.28 | 7 | Yes | Top 8 by both metrics |
| Widn AI Tower Sugarloaf | 0.808 | 8 | 1.28 | 6 | Yes | Top 8 by both metrics |
| Microsoft Translator | 0.804 | 20 | 1.22 | 2 | Yes | Top 8 by MetricX |
| AI21 Jamba Large 1.7 | 0.803 | 21 | 1.27 | 5 | Yes | Top 8 by MetricX |
| Elia Machine Translation System | 0.801 | 26 | 1.24 | 3 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Adjusted NMT | 1 | 0.818 | 1 | 1.14 | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 2 | 0.811 | 11 | 1.30 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Terra | Adjusted LLM | 3 | 0.811 | 8 | 1.29 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 4 | 0.810 | 4 | 1.27 | Top 8 by both metrics |
| Cohere Command A Translate | Adjusted LLM | 5 | 0.810 | 15 | 1.33 | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 6 | 0.809 | 17 | 1.34 | Top 8 by COMET-Kiwi |
| Amazon Translate | Off-the-shelf NMT | 7 | 0.809 | 7 | 1.28 | Top 8 by both metrics |
| Widn AI Tower Sugarloaf | Adjusted NMT | 8 | 0.808 | 6 | 1.28 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 9 | 0.807 | 14 | 1.33 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 10 | 0.807 | 20 | 1.35 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 11 | 0.807 | 29 | 1.41 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 12 | 0.807 | 13 | 1.32 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 13 | 0.807 | 27 | 1.41 | Not shortlisted |
| OpenAI GPT-5.5 | Adjusted LLM | 14 | 0.807 | 12 | 1.31 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 15 | 0.806 | 16 | 1.34 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 16 | 0.806 | 22 | 1.36 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 17 | 0.806 | 10 | 1.30 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 18 | 0.805 | 21 | 1.36 | Not shortlisted |
| Meta Llama 4 Maverick | Adjusted LLM | 19 | 0.804 | 28 | 1.41 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 20 | 0.804 | 2 | 1.22 | Top 8 by MetricX |
| AI21 Jamba Large 1.7 | Adjusted LLM | 21 | 0.803 | 5 | 1.27 | Top 8 by MetricX |
| Google Gemma 4 | Adjusted LLM | 22 | 0.803 | 19 | 1.35 | Not shortlisted |
| ModernMT Enterprise Edition | Adjusted NMT | 23 | 0.803 | 9 | 1.30 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 24 | 0.802 | 25 | 1.37 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 25 | 0.801 | 33 | 1.46 | Not shortlisted |
| Elia Machine Translation System | Off-the-shelf NMT | 26 | 0.801 | 3 | 1.24 | Top 8 by MetricX |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 27 | 0.801 | 23 | 1.36 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 28 | 0.800 | 38 | 1.67 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 29 | 0.800 | 31 | 1.43 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 30 | 0.799 | 24 | 1.36 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 31 | 0.796 | 26 | 1.40 | Not shortlisted |
| Google Translation LLM | Off-the-shelf NMT | 32 | 0.796 | 40 | 1.74 | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | Adjusted LLM | 33 | 0.795 | 32 | 1.43 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 34 | 0.793 | 30 | 1.42 | Not shortlisted |
| Tilde TildeOpen LLM engine | Adjusted NMT | 35 | 0.792 | 35 | 1.49 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 36 | 0.791 | 18 | 1.35 | Not shortlisted |
| SYSTRAN PNMT API | Adjusted NMT | 37 | 0.790 | 34 | 1.47 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 38 | 0.789 | 37 | 1.60 | Not shortlisted |
| Yandex Cloud Translate v2 | Adjusted NMT | 39 | 0.786 | 36 | 1.56 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 40 | 0.774 | 39 | 1.68 | Not shortlisted |
| Google Cloud Advanced Translation API | Adjusted NMT | 41 | 0.771 | 45 | 2.36 | Not shortlisted |
| iFlyTek MT | Off-the-shelf NMT | 42 | 0.771 | 41 | 1.87 | Not shortlisted |
| Tencent Cloud TMT | Off-the-shelf NMT | 43 | 0.765 | 42 | 1.98 | Not shortlisted |
| Naver Papago NMT Commercial | Off-the-shelf NMT | 44 | 0.764 | 43 | 2.03 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 45 | 0.759 | 44 | 2.05 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 46 | 0.751 | 46 | 2.38 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 47 | 0.699 | 47 | 3.32 | Not shortlisted |
| Widn AI Tower Anthill | Adjusted NMT | 48 | 0.485 | 48 | 10.14 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.807 | — | 1.36 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.806 | — | 1.33 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.803 | — | 1.40 | 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 48 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 shortlisted solutions differently. DeepL Next Gen, first on both metrics, misses general translation quality, tag handling and formatting, while xAI Grok 4.20 (reasoning), on the shortlist through COMET-Kiwi alone and 11th on MetricX, meets general translation quality. Microsoft Translator, second on MetricX and 20th on COMET-Kiwi, has 48.1 weighted errors per 1,000 words. The models nearest the cut are Google Gemini 3.5 Flash, 9th on COMET-Kiwi, and ModernMT Enterprise Edition, 9th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | xAI Grok 4.20 (reasoning), one-step LLM adjusted for Standard requirements | $1.01 |
| Standard | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | $3.74 |
| Advanced | OpenAI GPT-5.6 Terra, one-step with the gender-neutrality agent | $3.92 |
| Model | US dollars per million characters | Class |
|---|---|---|
| xAI Grok 4.20 (reasoning) | $1.01 | LLMs |
| AI21 Jamba Large 1.7 | $3.04 | LLMs |
| OpenAI GPT-5.6 Terra | $3.74 | LLMs |
| Google Gemini 3.1 Pro Preview | $3.87 | LLMs |
| Widn AI Tower Sugarloaf | $10 | LLM-based MT engines |
| Microsoft Translator | $10 | MT engines |
| Anthropic Claude Opus 4.8 | $14.06 | LLMs |
| Amazon Translate | $15 | MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.417 | LLMs |
| Anthropic Claude Sonnet 5 | 0.417 | LLMs |
| Anthropic Claude Fable 5 | 0.416 | LLMs |
| Mistral Medium 3.1 | 0.337 | LLMs |
| Mistral Large 3 | 0.326 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.309 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.308 | LLMs |
| MoonshotAI Kimi K2.5 | 0.292 | LLMs |
| AI21 Jamba Large 1.7 | 0.274 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.273 | Open-weight LLMs |
| Widn AI Tower Anthill | 0.272 | LLM-based MT engines |
| Widn AI Tower Sugarloaf | 0.271 | LLM-based MT engines |
| DeepSeek V3.2 | 0.268 | LLMs |
| z.ai GLM-5 | 0.255 | LLMs |
| Cohere Command A Translate | 0.248 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.242 | LLMs |
| Google Gemini 3.5 Flash | 0.242 | LLMs |
| Cohere Tiny Aya | 0.241 | LLMs |
| Google Gemma 4 | 0.241 | Open-weight LLMs |
| OpenAI GPT-4.1 | 0.238 | LLMs |
| OpenAI GPT-5.5 | 0.237 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.236 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.236 | LLMs |
| OpenAI GPT-5.6 Terra | 0.236 | LLMs |
| OpenAI GPT-5.6 Luna | 0.235 | LLMs |
| OpenAI GPT-5.6 Sol | 0.235 | LLMs |
| MiniMax M2.5 | 0.214 | LLMs |
With billing on tokens, the tokenizers produce a median 0.26 tokens per character of European Spanish, 1.1 times their count for English, and the tokenizer of Anthropic Claude Opus 4.8 produces 1.9 times as many tokens per character as the tokenizer of MiniMax M2.5.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | 17 |
| Standard | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | 17 |
| Advanced | OpenAI GPT-5.6 Terra, one-step with the gender-neutrality agent | 24 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| Pangeanic PangeaMT | 0.5 | MT engines |
| Google Cloud Advanced Translation API | 1 | MT engines |
| Translated Lara 2 | 1.1 | LLM-based MT engines |
| ModernMT Enterprise Edition | 1.2 | MT engines |
| Google Translation LLM | 1.2 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 1.3 | Open-weight LLMs |
| SYSTRAN PNMT API | 1.3 | MT engines |
| Amazon Translate | 1.5 | MT engines |
| DeepL Next Gen | 2 | LLM-based MT engines |
| Yandex Cloud Translate v2 | 2.9 | MT engines |
| Elia Machine Translation System | 3 | MT engines |
| Oracle Cloud Language Service | 4.1 | MT engines |
| Widn AI Tower Anthill | 7.8 | LLM-based MT engines |
| Tilde TildeOpen LLM engine | 10 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 13 | LLMs |
| OpenAI GPT-5.6 Luna | 13 | LLMs |
| Supertext LLM | 14 | LLM-based MT engines |
| OpenAI GPT-5.6 Terra | 17 | LLMs |
| OpenAI GPT-4.1 | 18 | LLMs |
| Cohere Tiny Aya | 18 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 19 | LLMs |
| Alibaba Machine Translation API | 22 | MT engines |
| Anthropic Claude Haiku 4.5 | 23 | LLMs |
| Cohere Command A Translate | 27 | LLMs |
| Mistral Medium 3.1 | 31 | LLMs |
| MiniMax M2.5 | 33 | LLMs |
| Anthropic Claude Sonnet 5 | 34 | LLMs |
| Mistral Large 3 | 37 | LLMs |
| DeepSeek V3.2 | 37 | LLMs |
| AI21 Jamba Large 1.7 | 40 | LLMs |
| OpenAI GPT-5.5 | 41 | LLMs |
| Meta Llama 4 Maverick | 41 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 41 | LLMs |
| Anthropic Claude Opus 4.8 | 43 | LLMs |
| Widn AI Tower Sugarloaf | 43 | LLM-based MT engines |
| Swiss AI Apertus-8B-2509 | 45 | Open-weight LLMs |
| Anthropic Claude Sonnet 4.6 | 51 | LLMs |
| iFlyTek MT | 52 | MT engines |
| SpeakLeash Bielik-11B-v3.0-Instruct | 54 | Open-weight LLMs |
| NiuTrans | 54 | MT engines |
| Tencent Cloud TMT | 56 | MT engines |
| xAI Grok 4.20 (reasoning) | 61 | LLMs |
| Google Gemini 3.1 Pro Preview | 73 | LLMs |
| Anthropic Claude Fable 5 | 73 | LLMs |
| Naver Papago NMT Commercial | 73 | MT engines |
| Alibaba Qwen3-235B-A22B | 92 | Open-weight LLMs |
| AppTek NMT | 95 | MT engines |
| Google Gemma 4 | 170 | Open-weight LLMs |
| z.ai GLM-5 | 189 | LLMs |
| MoonshotAI Kimi K2.5 | 237 | 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 European Spanish.
Nine of the ten real-time models in European Spanish are MT engines or LLM-based MT engines, and the other one is the open-weight OpenAI gpt-oss-120b, which did not make the shortlist. The shortlisted one-step LLM solutions take from 17 to 73 seconds per 1,000 words, OpenAI GPT-5.6 Terra the least and Google Gemini 3.1 Pro Preview the most.
| 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-5.6 Terra | one-step, with the gender-neutrality agent | 100% | ✓ | $3.92 | 24 |
| Google Gemini 3.1 Pro Preview | three-step, with the gender-neutrality agent | 99.1% | ✗ | $989 | 252 |
For gender neutrality in European Spanish we check neutral nouns and rephrasing, with paired forms only where they cannot be avoided. The gender-neutrality agent works with both of the two best solutions. With it, the three-step solution (translation + LQA + repair) on Google Gemini 3.1 Pro Preview costs $989 per million characters and takes 252 seconds per 1,000 words, against $3.92 and 24 seconds for the one-step solution on OpenAI GPT-5.6 Terra.
Both pairs of automatic metrics agree in European Spanish, and the three LQA judges agree on a median of 94.9% of the sentences of the shortlisted solutions, so the results on this page rest on measurement we can rely on.
Leave feedback