The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Greek 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. We test it in the 16 languages with a glossary, so it does not apply in Greek. |
| Tone of voice | The formal register (εσείς) 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 Greek, 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.
Four one-step solutions on LLMs meet every Standard requirement. The one on Google Gemini 3.1 Pro Preview costs the least, $5.66 per million characters, and has 16.7 total errors per 1,000 words. The one on Anthropic Claude Fable 5 costs the most, $39.73 per million characters, and has the fewest total errors, 13.3 per 1,000 words, in the least time, 95 seconds per 1,000 words. So Optimize for selects Google Gemini 3.1 Pro Preview for cost and Anthropic Claude Fable 5 for quality and latency.
Under the Standard requirements, a cost cap of at most $5.65 per million characters or a time limit of at most 94 seconds per 1,000 words removes all four solutions. Under the Advanced requirements, a cap below $47.29 per million characters or a limit below 249 seconds per 1,000 words removes both solutions. A limit on solution complexity that excludes the gender-neutrality agent keeps all four solutions for the Standard requirements and removes both for the Advanced requirements.
The step to Advanced leaves two solutions with the gender-neutrality agent. The two-step solution (translation + post-edit) on OpenAI GPT-5.5 costs $47.29 per million characters and takes 249 seconds per 1,000 words, against $13.18 and 98 for the one-step solution on the same model. The three-step solution (translation + LQA + repair) on Anthropic Claude Fable 5 costs $1,049 per million characters and takes 283 seconds per 1,000 words, against $39.73 and 95 for the one-step solution on the same model.
Against the one-step solutions on the same models, the two-step solution on OpenAI GPT-5.5 has 12.9 total errors per 1,000 words against 13.7, and the three-step solution on Anthropic Claude Fable 5 has 2.6 against 13.3. Weighted errors are 12.9 per 1,000 words against 10.9 for the solutions on OpenAI GPT-5.5, and 2.6 against 10.2 for those on Anthropic Claude Fable 5. So both have fewer total errors, at a higher cost and a longer time.
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
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 | ★×4 | one-step | Google Gemini 3.1 Pro Preview | 16.7 | $5.66 | 139 |
| Standard | Tier 1: simple solution + backup | ★×4 | one-step | Google Gemini 3.1 Pro Preview | 16.7 | $5.66 | 139 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | with the gender-neutrality agent | OpenAI GPT-5.5 | 12.9 | $47.29 | 249 |
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.
The same four one-step solutions meet every Basic and every Standard requirement. So the step from Basic to Standard keeps all four, one step is enough for both, and Optimize for selects the same solutions. The step to Advanced removes all four. Two solutions with the gender-neutrality agent meet it: a two-step solution (translation + post-edit) on OpenAI GPT-5.5 and a three-step solution (translation + LQA + repair) on Anthropic Claude Fable 5. Optimize for selects the first for cost and latency and the second for quality.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| xAI Grok 4.20 (reasoning) | 1.43 | 30.3 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.66 | 35.9 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 4.14 | 27.4 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.66 | 13.5 | Meets every Standard requirement |
| OpenAI GPT-5.6 Sol | 9.19 | 14.8 | Meets every Standard requirement |
| Microsoft Translator | 10 | 56 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 13.18 | 10.9 | Meets every Standard requirement |
| DeepL Next Gen | 27.5 | 23.1 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 39.73 | 10.2 | Meets every Standard requirement |
Three solutions that meet every Standard requirement form the front of cost and weighted errors: those on Google Gemini 3.1 Pro Preview, OpenAI GPT-5.5 and Anthropic Claude Fable 5. They cost $5.66, $13.18 and $39.73 per million characters, with 13.5, 10.9 and 10.2 weighted errors per 1,000 words. The fourth, on OpenAI GPT-5.6 Sol, is off the front: at $9.19 per million characters it has 14.8 weighted errors per 1,000 words, more than the cheaper solution on Google Gemini 3.1 Pro Preview.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 27.4 |
| Gender neutrality | 3.2 |
| Tone of voice | 0 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Of the one-step solutions that miss general translation quality, the one on DeepL Next Gen has the fewest weighted errors, 23.1 per 1,000 words, and five major errors. Two solutions come close on tag handling: in translations from Google Gemini 3.5 Flash, 99.8% of the tags are right, and from Cohere Command A Translate 99.2%. One step is enough for every Standard requirement, so two-step solutions (translation + post-edit) and three-step solutions (translation + LQA + repair) leave the tier of Greek unchanged.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Greek | Class |
|---|---|---|
| ModernMT Enterprise Edition | 64.6 | MT engines |
| Microsoft Translator | 56 | MT engines |
| Cohere Command A Translate | 38.2 | LLMs |
| OpenAI GPT-4.1 | 35.9 | LLMs |
| xAI Grok 4.20 (reasoning) | 30.3 | LLMs |
| Google Gemini 3.5 Flash | 27.4 | LLMs |
| DeepL Next Gen | 23.1 | LLM-based MT engines |
| OpenAI GPT-5.6 Sol | 14.8 | LLMs |
| Google Gemini 3.1 Pro Preview | 13.5 | LLMs |
| OpenAI GPT-5.5 | 10.9 | LLMs |
| Anthropic Claude Fable 5 | 10.2 | LLMs |
No translation into Greek: SpeakLeash Bielik-11B-v3.0-Instruct (wrong language, wrong script), Widn AI Tower Anthill (far too short, wrong language, wrong script), Widn AI Tower Vesuvius (far too short, wrong language, wrong script).
By Intento LQA, translations from the eight LLMs on the shortlist have 10.2 to 38.2 weighted errors per 1,000 words. Translations from the LLM-based MT engine DeepL Next Gen have 23.1 weighted errors per 1,000 words, and from the MT engines Microsoft Translator and ModernMT Enterprise Edition 56.0 and 64.6. The four solutions that meet general translation quality are all on LLMs. All eight shortlisted LLMs support Greek by the criteria we measure. None of the open-weight LLMs we got translations from are on the shortlist.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Opus 4.8 | 0.798 | 9 | 1.49 | 13 | No | Not shortlisted |
| Google Gemma 4 | 0.797 | 11 | 1.44 | 10 | No | Not shortlisted |
| DeepSeek V3.2 | 0.797 | 12 | 1.57 | 23 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.793 | 14 | 1.47 | 12 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.793 | 15 | 1.52 | 14 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.790 | 17 | 1.52 | 15 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.789 | 18 | 1.54 | 18 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.788 | 19 | 1.57 | 22 | No | Not shortlisted |
| Google Translation LLM | 0.788 | 20 | 1.79 | 28 | No | Not shortlisted |
| Mistral Large 3 | 0.787 | 21 | 1.46 | 11 | No | Not shortlisted |
| Cohere Tiny Aya | 0.787 | 22 | 1.54 | 16 | No | Not shortlisted |
| Translated Lara 2 | 0.787 | 23 | 1.64 | 25 | No | Not shortlisted |
| MiniMax M2.5 | 0.785 | 24 | 1.69 | 26 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.785 | 25 | 1.55 | 20 | No | Not shortlisted |
| z.ai GLM-5 | 0.785 | 26 | 1.54 | 19 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.784 | 27 | 1.59 | 24 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.782 | 28 | 1.81 | 29 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.776 | 29 | 1.74 | 27 | No | Not shortlisted |
| Amazon Translate | 0.775 | 30 | 1.90 | 30 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.770 | 31 | 2.06 | 32 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.762 | 32 | 2.43 | 35 | No | Not shortlisted |
| Supertext LLM | 0.761 | 33 | 2.02 | 31 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.760 | 34 | 2.27 | 34 | No | Not shortlisted |
| Baidu Translate API | 0.759 | 35 | 2.26 | 33 | No | Not shortlisted |
| AppTek NMT | 0.658 | 36 | 4.20 | 36 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.556 | 37 | 6.59 | 37 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.493 | 38 | 9.91 | 38 | No | Not shortlisted |
| Google Gemini 3.1 Pro Preview | 0.805 | 1 | 1.37 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | 0.802 | 2 | 1.24 | 1 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.802 | 3 | 1.42 | 9 | Yes | Top 8 by COMET-Kiwi |
| DeepL Next Gen | 0.801 | 4 | 1.31 | 2 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.801 | 5 | 1.39 | 5 | Yes | Top 8 by both metrics |
| Cohere Command A Translate | 0.800 | 6 | 1.41 | 7 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.800 | 7 | 1.34 | 3 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.800 | 8 | 1.40 | 6 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | 0.798 | 10 | 1.42 | 8 | Yes | Top 8 by MetricX |
| ModernMT Enterprise Edition | 0.793 | 13 | 1.54 | 17 | Yes | Real-time place |
| Microsoft Translator | 0.792 | 16 | 1.55 | 21 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 1 | 0.805 | 4 | 1.37 | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 2 | 0.802 | 1 | 1.24 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 3 | 0.802 | 9 | 1.42 | Top 8 by COMET-Kiwi |
| DeepL Next Gen | Off-the-shelf NMT | 4 | 0.801 | 2 | 1.31 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 5 | 0.801 | 5 | 1.39 | Top 8 by both metrics |
| Cohere Command A Translate | Adjusted LLM | 6 | 0.800 | 7 | 1.41 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 7 | 0.800 | 3 | 1.34 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 8 | 0.800 | 6 | 1.40 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 9 | 0.798 | 13 | 1.49 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 10 | 0.798 | 8 | 1.42 | Top 8 by MetricX |
| Google Gemma 4 | Adjusted LLM | 11 | 0.797 | 10 | 1.44 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 12 | 0.797 | 23 | 1.57 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 13 | 0.793 | 17 | 1.54 | Real-time place |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.793 | 12 | 1.47 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 15 | 0.793 | 14 | 1.52 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 16 | 0.792 | 21 | 1.55 | Real-time place |
| Mistral Medium 3.1 | Adjusted LLM | 17 | 0.790 | 15 | 1.52 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 18 | 0.789 | 18 | 1.54 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 19 | 0.788 | 22 | 1.57 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 20 | 0.788 | 28 | 1.79 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 21 | 0.787 | 11 | 1.46 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 22 | 0.787 | 16 | 1.54 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 23 | 0.787 | 25 | 1.64 | Not shortlisted |
| MiniMax M2.5 | Off-the-shelf LLM | 24 | 0.785 | 26 | 1.69 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 25 | 0.785 | 20 | 1.55 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 26 | 0.785 | 19 | 1.54 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 27 | 0.784 | 24 | 1.59 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 28 | 0.782 | 29 | 1.81 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 29 | 0.776 | 27 | 1.74 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 30 | 0.775 | 30 | 1.90 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 31 | 0.770 | 32 | 2.06 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 32 | 0.762 | 35 | 2.43 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 33 | 0.761 | 31 | 2.02 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 34 | 0.760 | 34 | 2.27 | Not shortlisted |
| Baidu Translate API | Off-the-shelf NMT | 35 | 0.759 | 33 | 2.26 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 36 | 0.658 | 36 | 4.20 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 37 | 0.556 | 37 | 6.59 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 38 | 0.493 | 38 | 9.91 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.800 | — | 1.37 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.798 | — | 1.51 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.790 | — | 1.47 | 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 38 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 automatic metrics and Intento LQA order the shortlist differently. Anthropic Claude Fable 5, 9th on MetricX, is on the list through COMET-Kiwi alone, and xAI Grok 4.20 (reasoning), 10th on COMET-Kiwi, through MetricX alone. Translations from them have 10.2 and 30.3 weighted errors per 1,000 words. Translations from DeepL Next Gen, 2nd on MetricX and 4th on COMET-Kiwi, have 23.1 weighted errors per 1,000 words. The models nearest the cut are Anthropic Claude Opus 4.8, 9th on COMET-Kiwi, and Google Gemma 4, 10th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $5.66 |
| Standard | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $5.66 |
| Advanced | OpenAI GPT-5.5, two-step with the gender-neutrality agent | $47.29 |
| Model | US dollars per million characters | Class |
|---|---|---|
| xAI Grok 4.20 (reasoning) | $1.43 | LLMs |
| OpenAI GPT-4.1 | $3.66 | LLMs |
| Google Gemini 3.5 Flash | $4.14 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.66 | LLMs |
| OpenAI GPT-5.6 Sol | $9.19 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $13.18 | LLMs |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $39.73 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Mistral Medium 3.1 | 0.859 | LLMs |
| Mistral Large 3 | 0.854 | LLMs |
| AI21 Jamba Large 1.7 | 0.774 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.736 | Open-weight LLMs |
| Anthropic Claude Fable 5 | 0.611 | LLMs |
| Anthropic Claude Opus 4.8 | 0.61 | LLMs |
| Anthropic Claude Sonnet 5 | 0.609 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.582 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.581 | LLMs |
| MiniMax M2.5 | 0.506 | LLMs |
| MoonshotAI Kimi K2.5 | 0.506 | LLMs |
| DeepSeek V3.2 | 0.43 | LLMs |
| z.ai GLM-5 | 0.38 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.376 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.374 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.352 | LLMs |
| Google Gemini 3.5 Flash | 0.351 | LLMs |
| Google Gemma 4 | 0.35 | Open-weight LLMs |
| OpenAI GPT-4.1 | 0.34 | LLMs |
| OpenAI GPT-5.6 Terra | 0.339 | LLMs |
| OpenAI GPT-5.5 | 0.338 | LLMs |
| OpenAI GPT-5.6 Sol | 0.337 | LLMs |
| OpenAI GPT-5.6 Luna | 0.334 | LLMs |
| Cohere Command A Translate | 0.294 | LLMs |
| Cohere Tiny Aya | 0.294 | LLMs |
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Anthropic Claude Fable 5, one-step LLM adjusted for Standard requirements | 95 |
| Standard | Anthropic Claude Fable 5, one-step LLM adjusted for Standard requirements | 95 |
| Advanced | OpenAI GPT-5.5, two-step with the gender-neutrality agent | 249 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.3 | MT engines |
| Google Cloud Advanced Translation API | 0.3 | MT engines |
| ModernMT Enterprise Edition | 0.6 | MT engines |
| Google Translation LLM | 1.2 | LLM-based MT engines |
| DeepL Next Gen | 1.3 | LLM-based MT engines |
| Amazon Translate | 1.6 | MT engines |
| Translated Lara 2 | 1.7 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.7 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.2 | MT engines |
| Oracle Cloud Language Service | 4.3 | MT engines |
| SYSTRAN PNMT API | 5.8 | MT engines |
| Baidu Translate API | 13 | MT engines |
| OpenAI GPT-5.6 Luna | 16 | LLMs |
| Google Gemini 3.5 Flash | 18 | LLMs |
| Supertext LLM | 20 | LLM-based MT engines |
| Cohere Tiny Aya | 23 | LLMs |
| OpenAI GPT-4.1 | 24 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 25 | LLMs |
| Alibaba Machine Translation API | 27 | MT engines |
| Cohere Command A Translate | 36 | LLMs |
| OpenAI GPT-5.6 Terra | 51 | LLMs |
| Mistral Medium 3.1 | 52 | LLMs |
| Swiss AI Apertus-8B-2509 | 54 | Open-weight LLMs |
| DeepSeek V3.2 | 56 | LLMs |
| Anthropic Claude Haiku 4.5 | 57 | LLMs |
| xAI Grok 4.20 (reasoning) | 60 | LLMs |
| MiniMax M2.5 | 62 | LLMs |
| Mistral Large 3 | 66 | LLMs |
| AI21 Jamba Large 1.7 | 79 | LLMs |
| Anthropic Claude Sonnet 5 | 92 | LLMs |
| AppTek NMT | 93 | MT engines |
| Anthropic Claude Fable 5 | 95 | LLMs |
| OpenAI GPT-5.5 | 98 | LLMs |
| Anthropic Claude Sonnet 4.6 | 109 | LLMs |
| Anthropic Claude Opus 4.8 | 121 | LLMs |
| Google Gemini 3.1 Pro Preview | 139 | LLMs |
| OpenAI GPT-5.6 Sol | 141 | LLMs |
| Google Gemma 4 | 190 | Open-weight LLMs |
| z.ai GLM-5 | 273 | LLMs |
| Alibaba Qwen3-235B-A22B | 275 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 287 | 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 Greek.
Eight of the nine real-time models in Greek are MT engines or LLM-based MT engines, and the ninth is the open-weight OpenAI gpt-oss-120b. The four solutions that meet every Standard requirement are on LLMs and take 95 to 141 seconds per 1,000 words. The same four meet every Basic requirement, so the time does not change from Basic to Standard. With the gender-neutrality agent, the two solutions that meet every Advanced requirement take 249 and 283 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 |
|---|---|---|---|---|---|
| Anthropic Claude Fable 5 | three-step, with the gender-neutrality agent | 100% | ✓ | $1,049 | 283 |
| OpenAI GPT-5.5 | two-step, with the gender-neutrality agent | 100% | ✓ | $47.29 | 249 |
For tone of voice in Greek we check the formal register, with εσείς as the form of address. In translations from DeepL Next Gen, 97.9% of the sentences are in the formal register. For gender neutrality we check neutral rephrasing and neutral nouns, with no default gender. None of the one-step solutions meet it, with a median 3.2 gender-neutrality errors per 1,000 words. With the gender-neutrality agent, 100% of the sentences about people are gender-neutral in translations from both of the two best solutions.
Leave feedback