The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Afrikaans 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 Afrikaans. |
| 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 Afrikaans, 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.
For the requirements of every level, Optimize for selects the same solution for quality, cost and latency: the three-step solution (translation + LQA + repair) on Anthropic Claude Fable 5, with the gender-neutrality agent for the Advanced requirements. A limit of one or two steps leaves solutions with at least 28.7 weighted errors per 1,000 words, on OpenAI GPT-5.5 with one step. A real-time limit (<2 sec) leaves three shortlisted solutions, and the one on Translated Lara 2 has the fewest total errors, 51.2 per 1,000 words.
Three three-step solutions on other models cost less and take less time, but have more weighted errors. Against the solution on Anthropic Claude Fable 5, the one on OpenAI GPT-5.6 Terra costs $492 instead of $562 per million characters and takes 152 instead of 203 seconds per 1,000 words. But it has more total errors, 21.8 instead of 19.1 per 1,000 words, with three major errors instead of none and weighted errors above the bar.
For the Advanced requirements, the gender-neutrality agent raises the cost of the three-step solution on Anthropic Claude Fable 5 from $562 to $968 per million characters. The time stays at 203 seconds per 1,000 words. Total errors rise from 19.1 to 19.4 per 1,000 words. Weighted errors rise from 18.8 to 19.4 per 1,000 words and major errors from none to one, both within the bars of general translation quality.
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Anthropic Claude Fable 5
three-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 3: one solution | ☆ | three-step | Anthropic Claude Fable 5 | 19.1 | $562 | 203 |
| Standard | Tier 3: one solution | ☆ | three-step | Anthropic Claude Fable 5 | 19.1 | $562 | 203 |
| Advanced | Tier 3: one solution | ☆ | with the gender-neutrality agent | Anthropic Claude Fable 5 | 19.4 | $968 | 203 |
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 step up from Basic to Standard keeps the one solution that meets the Basic requirements, as its translations have 100% of the sentences in the tone of voice and 100% of the tags in place. Terminology does not apply, as there is no glossary for Afrikaans. The step up to Advanced adds the gender-neutrality agent, and the solution with the gender-neutrality agent makes 100% of the sentences about people gender-neutral. The solution has 19.1 total errors per 1,000 words for the Basic and Standard requirements, and 19.4 with the agent.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.13 | 63.9 | Misses a Standard requirement |
| xAI Grok 4.20 (reasoning) | 1.09 | 49.8 | Misses a Standard requirement |
| MoonshotAI Kimi K2.5 | 1.21 | 64.9 | Misses a Standard requirement |
| Anthropic Claude Haiku 4.5 | 2.15 | 132.5 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.31 | 35.3 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 4.11 | 30.3 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 5.62 | 56 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 6.49 | 40.2 | Misses a Standard requirement |
| Microsoft Translator | 10 | 63.6 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 10.25 | 28.7 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 50.1 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 64.3 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 27.96 | 32 | Misses a Standard requirement |
The dearest one-step solutions have more weighted errors than several cheaper ones. The three dearest, on Anthropic Claude Fable 5, DeepL Next Gen and Translated Lara 2, cost $27.96, $27.50 and $24.99 per million characters and have 32.0, 64.3 and 50.1 weighted errors per 1,000 words. The one on OpenAI GPT-5.6 Terra costs $4.11 per million characters and has 30.3 weighted errors per 1,000 words. The one on OpenAI GPT-5.5 has the fewest, 28.7 weighted errors per 1,000 words, at $10.25 per million characters.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 50.1 |
| Tone of voice | 2.2 |
| Tag handling | 0.7 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Weighted errors are where most solutions fall short in general translation quality. A solution on Anthropic Claude Fable 5 has 32.0 weighted errors per 1,000 words with one step, 31.3 with two and 18.8 with three, under the bar. The other seven three-step solutions have 20.8 to 50.4 weighted errors per 1,000 words. Tags come close on two one-step solutions: translations from Translated Lara 2 have 99.8% of the tags in place and those from OpenAI GPT-5.5 99.6%.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Afrikaans | Class |
|---|---|---|
| Anthropic Claude Haiku 4.5 | 132.5 | LLMs |
| MoonshotAI Kimi K2.5 | 64.9 | LLMs |
| DeepL Next Gen | 64.3 | LLM-based MT engines |
| Google Gemma 4 | 63.9 | Open-weight LLMs |
| Microsoft Translator | 63.6 | MT engines |
| Anthropic Claude Sonnet 5 | 56 | LLMs |
| Translated Lara 2 | 50.1 | LLM-based MT engines |
| xAI Grok 4.20 (reasoning) | 49.8 | LLMs |
| Anthropic Claude Sonnet 4.6 | 40.2 | LLMs |
| Google Gemini 3.5 Flash | 35.3 | LLMs |
| Anthropic Claude Fable 5 | 32 | LLMs |
| OpenAI GPT-5.6 Terra | 30.3 | LLMs |
| OpenAI GPT-5.5 | 28.7 | LLMs |
The six one-step solutions with the fewest weighted errors are all on LLMs, from 28.7 to 49.8 per 1,000 words, and all six meet our language-support criteria. Translations from the best MT engine, the LLM-based Translated Lara 2, have 50.1 weighted errors per 1,000 words. The two off-the-shelf solutions, on Microsoft Translator and DeepL Next Gen, have 8.5% and 12.77% of the sentences in the tone of voice. The adjusted ones, on Translated Lara 2 and every LLM on the shortlist, have 97.9% to 100% of the sentences in the tone of voice.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Opus 4.8 | 0.747 | 11 | 1.51 | 12 | No | Not shortlisted |
| z.ai GLM-5 | 0.746 | 13 | 1.49 | 11 | No | Not shortlisted |
| Mistral Large 3 | 0.746 | 14 | 1.48 | 9 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.745 | 16 | 1.52 | 13 | No | Not shortlisted |
| Google Gemini 3.1 Pro Preview | 0.744 | 18 | 1.56 | 15 | No | Not shortlisted |
| DeepSeek V3.2 | 0.743 | 19 | 1.58 | 19 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.743 | 20 | 1.59 | 21 | No | Not shortlisted |
| Cohere Command A Translate | 0.741 | 21 | 1.56 | 17 | No | Not shortlisted |
| ALL-Lab | 0.740 | 22 | 1.59 | 20 | No | Not shortlisted |
| Google Translation LLM | 0.738 | 23 | 2.06 | 28 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.736 | 24 | 1.84 | 27 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.736 | 25 | 1.61 | 23 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.729 | 26 | 1.83 | 26 | No | Not shortlisted |
| MiniMax M2.5 | 0.728 | 27 | 1.67 | 24 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.725 | 28 | 1.74 | 25 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.723 | 29 | 2.06 | 29 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.721 | 30 | 2.41 | 31 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.709 | 31 | 2.33 | 30 | No | Not shortlisted |
| Amazon Translate | 0.693 | 32 | 2.61 | 32 | No | Not shortlisted |
| AppTek NMT | 0.666 | 33 | 3.39 | 33 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.637 | 34 | 3.40 | 34 | No | Not shortlisted |
| OpenAI GPT-5.6 Terra | 0.753 | 1 | 1.29 | 1 | Yes | Top 8 by both metrics |
| Anthropic Claude Haiku 4.5 | 0.752 | 2 | 1.57 | 18 | Yes | Top 8 by COMET-Kiwi |
| xAI Grok 4.20 (reasoning) | 0.752 | 3 | 1.42 | 6 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.751 | 4 | 1.42 | 4 | Yes | Top 8 by both metrics |
| Google Gemma 4 | 0.750 | 5 | 1.47 | 8 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.750 | 6 | 1.48 | 10 | Yes | Top 8 by COMET-Kiwi |
| Translated Lara 2 | 0.750 | 7 | 1.60 | 22 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Fable 5 | 0.749 | 8 | 1.55 | 14 | Yes | Top 8 by COMET-Kiwi |
| Microsoft Translator | 0.748 | 9 | 1.56 | 16 | Yes | Real-time place |
| OpenAI GPT-5.5 | 0.748 | 10 | 1.41 | 3 | Yes | Top 8 by MetricX |
| Anthropic Claude Sonnet 4.6 | 0.747 | 12 | 1.42 | 5 | Yes | Top 8 by MetricX |
| Google Gemini 3.5 Flash | 0.745 | 15 | 1.33 | 2 | Yes | Top 8 by MetricX |
| MoonshotAI Kimi K2.5 | 0.744 | 17 | 1.45 | 7 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Terra | Adjusted LLM | 1 | 0.753 | 1 | 1.29 | Top 8 by both metrics |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 2 | 0.752 | 18 | 1.57 | Top 8 by COMET-Kiwi |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 3 | 0.752 | 6 | 1.42 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 4 | 0.751 | 4 | 1.42 | Top 8 by both metrics |
| Google Gemma 4 | Adjusted LLM | 5 | 0.750 | 8 | 1.47 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 6 | 0.750 | 10 | 1.48 | Top 8 by COMET-Kiwi |
| Translated Lara 2 | Adjusted NMT | 7 | 0.750 | 22 | 1.60 | Top 8 by COMET-Kiwi |
| Anthropic Claude Fable 5 | Adjusted LLM | 8 | 0.749 | 14 | 1.55 | Top 8 by COMET-Kiwi |
| Microsoft Translator | Off-the-shelf NMT | 9 | 0.748 | 16 | 1.56 | Real-time place |
| OpenAI GPT-5.5 | Adjusted LLM | 10 | 0.748 | 3 | 1.41 | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 11 | 0.747 | 12 | 1.51 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 12 | 0.747 | 5 | 1.42 | Top 8 by MetricX |
| z.ai GLM-5 | Adjusted LLM | 13 | 0.746 | 11 | 1.49 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 14 | 0.746 | 9 | 1.48 | Not shortlisted |
| Google Gemini 3.5 Flash | Adjusted LLM | 15 | 0.745 | 2 | 1.33 | Top 8 by MetricX |
| OpenAI GPT-4.1 | Adjusted LLM | 16 | 0.745 | 13 | 1.52 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 17 | 0.744 | 7 | 1.45 | Top 8 by MetricX |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 18 | 0.744 | 15 | 1.56 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 19 | 0.743 | 19 | 1.58 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 20 | 0.743 | 21 | 1.59 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 21 | 0.741 | 17 | 1.56 | Not shortlisted |
| ALL-Lab | Off-the-shelf NMT | 22 | 0.740 | 20 | 1.59 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 23 | 0.738 | 28 | 2.06 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 24 | 0.736 | 27 | 1.84 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 25 | 0.736 | 23 | 1.61 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 26 | 0.729 | 26 | 1.83 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 27 | 0.728 | 24 | 1.67 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 28 | 0.725 | 25 | 1.74 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 29 | 0.723 | 29 | 2.06 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 30 | 0.721 | 31 | 2.41 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 31 | 0.709 | 30 | 2.33 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 32 | 0.693 | 32 | 2.61 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 33 | 0.666 | 33 | 3.39 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 34 | 0.637 | 34 | 3.40 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.751 | — | 1.35 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.751 | — | 1.51 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.736 | — | 1.57 | 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 34 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.
Translations from Anthropic Claude Haiku 4.5, second on COMET-Kiwi, have 132.5 weighted errors per 1,000 words, the most of the 13 shortlisted solutions. The fewest, 28.7 weighted errors per 1,000 words, come from OpenAI GPT-5.5, on the list through MetricX alone. Anthropic Claude Fable 5, the model of the solutions that meet the levels, is 8th on COMET-Kiwi, the last place in the top 8. The tail models nearest the cut are Mistral Large 3, 9th on MetricX, and Anthropic Claude Opus 4.8, 11th on COMET-Kiwi.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Anthropic Claude Fable 5, three-step | $562 |
| Standard | Anthropic Claude Fable 5, three-step | $562 |
| Advanced | Anthropic Claude Fable 5, three-step with the gender-neutrality agent | $968 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.13 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | $1.09 | LLMs |
| MoonshotAI Kimi K2.5 | $1.21 | LLMs |
| Anthropic Claude Haiku 4.5 | $2.15 | LLMs |
| Google Gemini 3.5 Flash | $3.31 | LLMs |
| OpenAI GPT-5.6 Terra | $4.11 | LLMs |
| Anthropic Claude Sonnet 5 | $5.62 | LLMs |
| Anthropic Claude Sonnet 4.6 | $6.49 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $10.25 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $27.96 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Sonnet 5 | 0.468 | LLMs |
| Anthropic Claude Opus 4.8 | 0.467 | LLMs |
| Anthropic Claude Fable 5 | 0.466 | LLMs |
| Mistral Medium 3.1 | 0.379 | LLMs |
| Mistral Large 3 | 0.373 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.366 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.361 | LLMs |
| AI21 Jamba Large 1.7 | 0.343 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.337 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 0.337 | LLMs |
| DeepSeek V3.2 | 0.327 | LLMs |
| z.ai GLM-5 | 0.324 | LLMs |
| Cohere Command A Translate | 0.314 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.309 | LLMs |
| Google Gemini 3.5 Flash | 0.309 | LLMs |
| Google Gemma 4 | 0.307 | Open-weight LLMs |
| MiniMax M2.5 | 0.301 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.298 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.298 | LLMs |
| OpenAI GPT-4.1 | 0.288 | LLMs |
| OpenAI GPT-5.5 | 0.288 | LLMs |
| OpenAI GPT-5.6 Terra | 0.288 | LLMs |
| OpenAI GPT-5.6 Luna | 0.287 | LLMs |
| OpenAI GPT-5.6 Sol | 0.286 | LLMs |
A character of Afrikaans comes to a median 0.32 tokens across the tokenizers of 24 models, 1.3 times the figure for English. The three-step solutions cost $492 to $629 per million characters, against $0.13 to $27.96 for the one-step solutions. Meeting the Basic and Standard requirements costs the same, $562 per million characters, as one solution meets both. Meeting the Advanced requirements costs $968 per million characters, with the gender-neutrality agent.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Anthropic Claude Fable 5, three-step | 203 |
| Standard | Anthropic Claude Fable 5, three-step | 203 |
| Advanced | Anthropic Claude Fable 5, three-step with the gender-neutrality agent | 203 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.1 | MT engines |
| Microsoft Translator | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.4 | MT engines |
| Google Translation LLM | 0.5 | LLM-based MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| Amazon Translate | 1.2 | MT engines |
| Translated Lara 2 | 1.3 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 1.8 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.6 | MT engines |
| OpenAI GPT-5.6 Luna | 14 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 19 | LLMs |
| OpenAI GPT-4.1 | 23 | LLMs |
| Alibaba Machine Translation API | 26 | MT engines |
| OpenAI GPT-5.6 Terra | 28 | LLMs |
| ALL-Lab | 28 | MT engines |
| Anthropic Claude Haiku 4.5 | 36 | LLMs |
| Mistral Medium 3.1 | 37 | LLMs |
| MiniMax M2.5 | 40 | LLMs |
| AI21 Jamba Large 1.7 | 46 | LLMs |
| Cohere Command A Translate | 47 | LLMs |
| xAI Grok 4.20 (reasoning) | 53 | LLMs |
| Mistral Large 3 | 53 | LLMs |
| Anthropic Claude Sonnet 5 | 58 | LLMs |
| OpenAI GPT-5.6 Sol | 62 | LLMs |
| Anthropic Claude Fable 5 | 63 | LLMs |
| DeepSeek V3.2 | 66 | LLMs |
| Anthropic Claude Sonnet 4.6 | 70 | LLMs |
| Google Gemini 3.5 Flash | 70 | LLMs |
| Anthropic Claude Opus 4.8 | 73 | LLMs |
| Google Gemini 3.1 Pro Preview | 77 | LLMs |
| OpenAI GPT-5.5 | 82 | LLMs |
| Swiss AI Apertus-8B-2509 | 85 | Open-weight LLMs |
| AppTek NMT | 126 | MT engines |
| Alibaba Qwen3-235B-A22B | 133 | Open-weight LLMs |
| Google Gemma 4 | 161 | Open-weight LLMs |
| z.ai GLM-5 | 197 | LLMs |
| MoonshotAI Kimi K2.5 | 216 | 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 Afrikaans.
Seven of the eight real-time models in Afrikaans are MT engines or LLM-based MT engines, and the eighth is the open-weight OpenAI gpt-oss-120b, outside the shortlist. Translations from the three on the shortlist, Microsoft Translator, DeepL Next Gen and Translated Lara 2, have 63.6, 64.3 and 50.1 weighted errors per 1,000 words. None of them meet formatting or tag handling. The solutions that meet the levels take 203 seconds per 1,000 words, with or without the gender-neutrality agent.
| 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% | ✓ | $968 | 203 |
| OpenAI GPT-5.6 Terra | three-step | 100% | ✗ | $492 | 152 |
The tone of voice we check in Afrikaans is the formal register, with "u" as the form of address. For gender neutrality we check rephrasing, plural forms and neutral nouns, and we allow no default gender. The other options in the language are generic "hy" and "hy of sy". Both of the two best solutions make 100% of the sentences about people gender-neutral, the one on OpenAI GPT-5.6 Terra without the agent. Its 21.8 weighted errors per 1,000 words are more than general translation quality allows.
Both metric pairs agree above the threshold, and neither result is borderline. The three LQA judges agree on a median of 97.2% of the sentences across the 13 shortlisted solutions. So the metric rankings and the Intento LQA results on this page are well grounded.
Leave feedback