The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Belarusian 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 (вы) 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 Belarusian, 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 solutions meet every Standard requirement in Belarusian, and none leads on errors, cost and time. For quality, three tie at 22.2 total errors per 1,000 words: the three-step solution (translation + LQA + repair) on z.ai GLM-5, at $600 per million characters, and the one-step solutions on Google Gemini 3.1 Pro Preview and Anthropic Claude Fable 5. For cost, Optimize for selects the one on Google Gemini 3.1 Pro Preview, at $5.41 per million characters, and for latency the one on Anthropic Claude Fable 5, at 96 seconds per 1,000 words.
A cost cap under $5.41 per million characters or a time limit of at most 95 seconds per 1,000 words removes all four solutions that meet every Standard requirement in Belarusian. A limit of one step drops the three-step solution (translation + LQA + repair) on z.ai GLM-5, and two of the three left have the same 22.2 total errors per 1,000 words. Under the Advanced requirements, a cost cap of at most $77.03 per million characters, a time limit under 181 seconds per 1,000 words or a limit of one step removes both solutions that meet them.
For the Advanced requirements, the gender-neutrality agent raises the cost of the three-step solution (translation + LQA + repair) on OpenAI GPT-5.6 Sol from $608 to $982 per million characters. Its total errors fall from 10.4 to 7.9 per 1,000 words, and its time is 238 seconds per 1,000 words against 259. With the agent, the total errors of the two-step solution (translation + post-edit) on Anthropic Claude Fable 5 fall from 19.5 to 15.5 per 1,000 words, and its time is 181 seconds per 1,000 words against 183.
z.ai GLM-5
three-step
Tied with Anthropic Claude Fable 5 (one-step LLM adjusted for Standard requirements) and Google Gemini 3.1 Pro Preview (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
z.ai GLM-5
three-step
Tied with Anthropic Claude Fable 5 (one-step LLM adjusted for Standard requirements) and Google Gemini 3.1 Pro Preview (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
OpenAI GPT-5.6 Sol
three-step with the gender-neutrality agent
Anthropic Claude Fable 5
two-step with the gender-neutrality agent
Anthropic Claude Fable 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 | Google Gemini 3.1 Pro Preview | 22.2 | $5.41 | 97 |
| Standard | Tier 1: simple solution + backup | ★★★ ☆ | one-step | Google Gemini 3.1 Pro Preview | 22.2 | $5.41 | 97 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | with the gender-neutrality agent | Anthropic Claude Fable 5 | 15.5 | $77.04 | 181 |
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.
In Belarusian the same four solutions meet every Basic and every Standard requirement, so Optimize for selects the same solutions at both levels. The Advanced requirements remove all four and leave two solutions with the gender-neutrality agent. Optimize for then selects the two-step solution (translation + post-edit) on Anthropic Claude Fable 5 for cost and latency, and the three-step solution (translation + LQA + repair) on OpenAI GPT-5.6 Sol for quality.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| z.ai GLM-5 | 2.4 | 32 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.39 | 48.8 | Misses a Standard requirement |
| Yandex Cloud Translate v2 | 3.94 | 143.7 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 4.05 | 24.1 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.41 | 19.8 | Meets every Standard requirement |
| OpenAI GPT-5.6 Sol | 8.45 | 19.8 | Meets every Standard requirement |
| Anthropic Claude Sonnet 4.6 | 8.95 | 40.9 | Misses a Standard requirement |
| Google Translation LLM | 21.09 | 70.5 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 64.3 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 56.4 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 32.09 | 19.8 | Meets every Standard requirement |
Among the one-step solutions in Belarusian, those on the three LLM-based MT engines cost $21.09 to $27.50 per million characters, more than every LLM solution but the one on Anthropic Claude Fable 5. They have 56.4 to 70.5 weighted errors per 1,000 words, more than any LLM solution. The cheapest, on z.ai GLM-5 and OpenAI GPT-4.1, cost $2.40 and $3.39 per million characters and have 32.0 and 48.8 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 40.9 |
| Gender neutrality | 2.8 |
| Terminology | 0 |
| Tone of voice | 0 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
In Belarusian, the one-step solutions that miss general translation quality have 24.1 to 143.7 weighted errors per 1,000 words. On z.ai GLM-5, the one-step solution has 32.0 weighted errors per 1,000 words, the two-step solution (translation + post-edit) 32.3 and the three-step solution (translation + LQA + repair) 19.1, under the bar. In one step, 100% of the glossary terms are right in the translations from six out of seven shortlisted LLMs, and 45.9% to 97.5% in those from the four MT engines, three of them LLM-based.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Belarusian | Class |
|---|---|---|
| Yandex Cloud Translate v2 | 143.7 | MT engines |
| Google Translation LLM | 70.5 | LLM-based MT engines |
| Translated Lara 2 | 64.3 | LLM-based MT engines |
| DeepL Next Gen | 56.4 | LLM-based MT engines |
| OpenAI GPT-4.1 | 48.8 | LLMs |
| Anthropic Claude Sonnet 4.6 | 40.9 | LLMs |
| z.ai GLM-5 | 32 | LLMs |
| Google Gemini 3.5 Flash | 24.1 | LLMs |
| Anthropic Claude Fable 5 | 19.8 | LLMs |
| OpenAI GPT-5.6 Sol | 19.8 | LLMs |
| Google Gemini 3.1 Pro Preview | 19.8 | LLMs |
No translation into Belarusian: SpeakLeash Bielik-11B-v3.0-Instruct (wrong language, wrong script).
In one step, translations from the seven shortlisted LLMs have 19.8 to 48.8 weighted errors per 1,000 words in Belarusian, those from the three LLM-based MT engines 56.4 to 70.5, and those from Yandex Cloud Translate v2, the one other MT engine, 143.7. Among them, only translations from Google Translation LLM and Yandex Cloud Translate v2 have critical errors, 9 and 1. All seven LLMs support Belarusian by our measured criteria. The output of SpeakLeash Bielik-11B-v3.0-Instruct came back in the wrong language and script.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.5 | 0.777 | 9 | 1.55 | 10 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.776 | 11 | 1.56 | 11 | No | Not shortlisted |
| Mistral Large 3 | 0.776 | 12 | 1.57 | 12 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.772 | 13 | 1.71 | 15 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.770 | 14 | 1.59 | 13 | No | Not shortlisted |
| DeepSeek V3.2 | 0.769 | 15 | 1.78 | 16 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.767 | 16 | 1.80 | 17 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.766 | 17 | 1.89 | 19 | No | Not shortlisted |
| Google Gemma 4 | 0.765 | 18 | 1.92 | 20 | No | Not shortlisted |
| Cohere Command A Translate | 0.753 | 21 | 1.96 | 21 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.748 | 22 | 2.06 | 22 | No | Not shortlisted |
| MiniMax M2.5 | 0.744 | 23 | 2.28 | 23 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.728 | 24 | 2.44 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.727 | 25 | 2.63 | 26 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.717 | 26 | 2.42 | 24 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.713 | 27 | 3.12 | 27 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.632 | 28 | 4.79 | 28 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.463 | 29 | 8.76 | 29 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.285 | 30 | 13.84 | 30 | No | Not shortlisted |
| AppTek NMT | 0.192 | 31 | 16.96 | 31 | No | Not shortlisted |
| OpenAI GPT-5.6 Sol | 0.786 | 1 | 1.33 | 1 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.782 | 2 | 1.42 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.781 | 3 | 1.52 | 8 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.780 | 4 | 1.46 | 5 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.780 | 5 | 1.49 | 6 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | 0.779 | 6 | 1.36 | 2 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.779 | 7 | 1.53 | 9 | Yes | Top 8 by COMET-Kiwi |
| Google Translation LLM | 0.778 | 8 | 1.68 | 14 | Yes | Top 8 by COMET-Kiwi |
| DeepL Next Gen | 0.777 | 10 | 1.37 | 3 | Yes | Top 8 by MetricX |
| Yandex Cloud Translate v2 | 0.760 | 19 | 1.51 | 7 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.760 | 20 | 1.89 | 18 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Sol | Adjusted LLM | 1 | 0.786 | 1 | 1.33 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 2 | 0.782 | 4 | 1.42 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 3 | 0.781 | 8 | 1.52 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 4 | 0.780 | 5 | 1.46 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 5 | 0.780 | 6 | 1.49 | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 6 | 0.779 | 2 | 1.36 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 7 | 0.779 | 9 | 1.53 | Top 8 by COMET-Kiwi |
| Google Translation LLM | Off-the-shelf NMT | 8 | 0.778 | 14 | 1.68 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.5 | Adjusted LLM | 9 | 0.777 | 10 | 1.55 | Not shortlisted |
| DeepL Next Gen | Off-the-shelf NMT | 10 | 0.777 | 3 | 1.37 | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 11 | 0.776 | 11 | 1.56 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 12 | 0.776 | 12 | 1.57 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 13 | 0.772 | 15 | 1.71 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.770 | 13 | 1.59 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 15 | 0.769 | 16 | 1.78 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 16 | 0.767 | 17 | 1.80 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 17 | 0.766 | 19 | 1.89 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 18 | 0.765 | 20 | 1.92 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 19 | 0.760 | 7 | 1.51 | Top 8 by MetricX |
| Translated Lara 2 | Adjusted NMT | 20 | 0.760 | 18 | 1.89 | Real-time place |
| Cohere Command A Translate | Adjusted LLM | 21 | 0.753 | 21 | 1.96 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 22 | 0.748 | 22 | 2.06 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 23 | 0.744 | 23 | 2.28 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 24 | 0.728 | 25 | 2.44 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 25 | 0.727 | 26 | 2.63 | Not shortlisted |
| ModernMT Enterprise Edition | Adjusted NMT | 26 | 0.717 | 24 | 2.42 | Not shortlisted |
| Google Cloud Advanced Translation API | Adjusted NMT | 27 | 0.713 | 27 | 3.12 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 28 | 0.632 | 28 | 4.79 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 29 | 0.463 | 29 | 8.76 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 30 | 0.285 | 30 | 13.84 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 31 | 0.192 | 31 | 16.96 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.780 | — | 1.52 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.781 | — | 1.57 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.765 | — | 1.78 | 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 31 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.
MetricX ranks Anthropic Claude Sonnet 4.6 second and DeepL Next Gen third in Belarusian. By Intento LQA, their one-step solutions have 40.9 and 56.4 weighted errors per 1,000 words, against 19.8 for Google Gemini 3.1 Pro Preview, fifth on COMET-Kiwi and sixth on MetricX. COMET-Kiwi alone brought z.ai GLM-5 onto the list, and its three-step solution (translation + LQA + repair) meets every Standard requirement. Nearest the cut in the tail is OpenAI GPT-5.5, ninth on COMET-Kiwi.
| 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 | $5.41 |
| Standard | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $5.41 |
| Advanced | Anthropic Claude Fable 5, two-step with the gender-neutrality agent | $77.04 |
| Model | US dollars per million characters | Class |
|---|---|---|
| z.ai GLM-5 | $2.4 | LLMs |
| OpenAI GPT-4.1 | $3.39 | LLMs |
| Yandex Cloud Translate v2 | $3.94 | MT engines |
| Google Gemini 3.5 Flash | $4.05 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.41 | LLMs |
| OpenAI GPT-5.6 Sol | $8.45 | LLMs |
| Anthropic Claude Sonnet 4.6 | $8.95 | LLMs |
| Google Translation LLM | $21.09 | LLM-based MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $32.09 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Sonnet 5 | 0.528 | LLMs |
| Anthropic Claude Opus 4.8 | 0.524 | LLMs |
| Anthropic Claude Fable 5 | 0.523 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.495 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.494 | Open-weight LLMs |
| Anthropic Claude Haiku 4.5 | 0.494 | LLMs |
| Mistral Medium 3.1 | 0.494 | LLMs |
| Mistral Large 3 | 0.492 | LLMs |
| DeepSeek V3.2 | 0.44 | LLMs |
| MoonshotAI Kimi K2.5 | 0.438 | LLMs |
| z.ai GLM-5 | 0.423 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.415 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.412 | LLMs |
| MiniMax M2.5 | 0.389 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.374 | LLMs |
| Google Gemini 3.5 Flash | 0.374 | LLMs |
| Google Gemma 4 | 0.372 | Open-weight LLMs |
| Cohere Command A Translate | 0.366 | LLMs |
| OpenAI GPT-5.6 Sol | 0.34 | LLMs |
| OpenAI GPT-5.6 Terra | 0.339 | LLMs |
| OpenAI GPT-5.6 Luna | 0.338 | LLMs |
| OpenAI GPT-5.5 | 0.337 | LLMs |
| OpenAI GPT-4.1 | 0.336 | LLMs |
| AI21 Jamba Large 1.7 | 0.322 | LLMs |
With billing on tokens, the tokenizers produce a median 0.41 tokens per character of a translation into Belarusian, 1.6 times their count for English. On one model, the cost follows the steps. The solutions on z.ai GLM-5 cost $2.40 per million characters in one step, $5.48 in two steps (translation + post-edit) and $600 in three steps (translation + LQA + repair). Those on OpenAI GPT-5.6 Sol cost $8.45 per million characters in one step and $608 in three steps.
| 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 | 96 |
| Standard | Anthropic Claude Fable 5, one-step LLM adjusted for Standard requirements | 96 |
| Advanced | Anthropic Claude Fable 5, two-step with the gender-neutrality agent | 181 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Translation LLM | 0.4 | LLM-based MT engines |
| Google Cloud Advanced Translation API | 0.5 | MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| ModernMT Enterprise Edition | 1 | MT engines |
| Translated Lara 2 | 1.8 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.7 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 4.8 | MT engines |
| Google Gemini 3.5 Flash | 16 | LLMs |
| OpenAI GPT-5.6 Luna | 17 | LLMs |
| OpenAI GPT-4.1 | 24 | LLMs |
| Alibaba Machine Translation API | 27 | MT engines |
| xAI Grok 4.20 (non-reasoning) | 31 | LLMs |
| OpenAI GPT-5.6 Terra | 45 | LLMs |
| DeepSeek V3.2 | 49 | LLMs |
| Mistral Medium 3.1 | 53 | LLMs |
| Anthropic Claude Haiku 4.5 | 55 | LLMs |
| Cohere Command A Translate | 57 | LLMs |
| MiniMax M2.5 | 58 | LLMs |
| AI21 Jamba Large 1.7 | 71 | LLMs |
| xAI Grok 4.20 (reasoning) | 77 | LLMs |
| Mistral Large 3 | 78 | LLMs |
| Swiss AI Apertus-8B-2509 | 86 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 90 | LLMs |
| AppTek NMT | 91 | MT engines |
| Anthropic Claude Fable 5 | 96 | LLMs |
| Google Gemini 3.1 Pro Preview | 97 | LLMs |
| OpenAI GPT-5.5 | 108 | LLMs |
| OpenAI GPT-5.6 Sol | 114 | LLMs |
| Anthropic Claude Sonnet 4.6 | 114 | LLMs |
| Anthropic Claude Opus 4.8 | 123 | LLMs |
| Alibaba Qwen3-235B-A22B | 163 | Open-weight LLMs |
| Google Gemma 4 | 220 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 251 | LLMs |
| z.ai GLM-5 | 267 | 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 Belarusian.
The six real-time models in Belarusian are two MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. Three are on the shortlist, DeepL Next Gen, Google Translation LLM and Translated Lara 2, and their translations have 56.4 to 70.5 weighted errors per 1,000 words. None of the three meet general translation quality or terminology. From the Standard to the Advanced requirements, the time of the fastest solution goes from 96 to 181 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-5.6 Sol | three-step, with the gender-neutrality agent | 100% | ✓ | $982 | 238 |
| Anthropic Claude Fable 5 | two-step, with the gender-neutrality agent | 100% | ✓ | $77.04 | 181 |
For Belarusian we check the formal register, with вы, and gender-neutral wording through neutral rephrasing, plurals and impersonal forms, with no default gender. The one-step solutions have a median 2.8 gender-neutrality errors per 1,000 words. The two best solutions, the two-step one (translation + post-edit) on Anthropic Claude Fable 5 and the three-step one (translation + LQA + repair) on OpenAI GPT-5.6 Sol, work with the gender-neutrality agent. In their translations, 100% of the sentences about people are gender-neutral.
For the 11 shortlisted solutions in Belarusian, the median share of sentences on which the three LQA judges agree is 94.9%. With agreement that high, the Intento LQA error counts of these 11 solutions are reliable.
Leave feedback