The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Uyghur 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 Uyghur. |
| Tone of voice | The formal register (سىز) 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. |
| Metric pair | Metrics | Agreement in Uyghur | 90% band | Minimum | Result |
|---|---|---|---|---|---|
| Similarity | LaBSE, Cohere embed v3 | 0.435 | 0.421 to 0.451 | 0.461 | Under the minimum |
| Quality estimation | COMET-Kiwi, MetricX | — | — | 0.534 | Cannot be applied |
Pick a quality level, or your own requirements, bars and constraints, to see the shortlisted solutions that meet them in Uyghur, with a column for each requirement you select. Under Optimize for, choose which of them comes first: the one with the lowest cost or the shortest time. Click a column heading to sort by it.
We did not shortlist solutions in Uyghur, so the explorer has none: why.
Of the 27 models we got translations into Uyghur from, a limit of 2 seconds per text leaves the three MT engines. A limit of 3.6 seconds per 1,000 words adds Translated Lara 2 and OpenAI gpt-oss-120b. A limit of 63 seconds per 1,000 words adds OpenAI GPT-5.6 Luna, xAI Grok 4.20 (non-reasoning), DeepSeek V3.2 and OpenAI GPT-4.1. These limits rest on time alone.
Translations from the open-weight Google Gemma 4, the cheapest with a cost, take 312 seconds per 1,000 words. Those from OpenAI GPT-5.6 Luna cost $0.63 per million characters, a lower bound, and take 22 seconds per 1,000 words. The real-time Microsoft Translator and Google Cloud Advanced Translation API take 0.2 seconds per 1,000 words, and their translations cost $10 and $20 per million characters.
We cannot assess general translation quality in Uyghur: how confident we are.
We cannot assess general translation quality in Uyghur: how confident we are.
We cannot assess general translation quality in Uyghur: how confident we are.
No translation into Uyghur: AI21 Jamba Large 1.7 (garbage or near-empty, wrong language, wrong script), Swiss AI Apertus-8B-2509 (wrong language, wrong script, garbage or near-empty).
The rest of the page does not rest on a ranking: the 27 models we got translations from, the cost of the 24 models with a cost, the time of the 27 models we timed, the tokens per character from the tokenizers of 22 models, the length of a translation and the tone of voice and gender-neutrality requirements of Uyghur.
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.21 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | $0.63 | LLMs |
| MiniMax M2.5 | $1.11 | LLMs |
| Mistral Large 3 | $1.55 | LLMs |
| DeepSeek V3.2 | $1.62 | LLMs |
| Mistral Medium 3.1 | $1.94 | LLMs |
| xAI Grok 4.20 (reasoning) | $1.97 | LLMs |
| xAI Grok 4.20 (non-reasoning) | $1.98 | LLMs |
| MoonshotAI Kimi K2.5 | $2.25 | LLMs |
| z.ai GLM-5 | $3.53 | LLMs |
| Anthropic Claude Haiku 4.5 | $3.95 | LLMs |
| OpenAI GPT-4.1 | $4.02 | LLMs |
| Google Gemini 3.5 Flash | $5.63 | LLMs |
| OpenAI GPT-5.6 Terra | $6.39 | LLMs |
| Google Gemini 3.1 Pro Preview | $7.34 | LLMs |
| Anthropic Claude Sonnet 5 | $8.75 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.6 Sol | $10.87 | LLMs |
| Anthropic Claude Sonnet 4.6 | $13.4 | LLMs |
| OpenAI GPT-5.5 | $15.93 | LLMs |
| Google Cloud Advanced Translation API | $20 | MT engines |
| Anthropic Claude Opus 4.8 | $24.6 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $49.09 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Mistral Large 3 | 0.902 | LLMs |
| Mistral Medium 3.1 | 0.896 | LLMs |
| Anthropic Claude Opus 4.8 | 0.834 | LLMs |
| Anthropic Claude Fable 5 | 0.83 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.799 | LLMs |
| Anthropic Claude Sonnet 5 | 0.75 | LLMs |
| Cohere Command A Translate | 0.731 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.719 | LLMs |
| z.ai GLM-5 | 0.65 | LLMs |
| MiniMax M2.5 | 0.649 | LLMs |
| MoonshotAI Kimi K2.5 | 0.649 | LLMs |
| DeepSeek V3.2 | 0.637 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.63 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.627 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.507 | LLMs |
| Google Gemini 3.5 Flash | 0.507 | LLMs |
| Google Gemma 4 | 0.504 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | 0.441 | LLMs |
| OpenAI GPT-5.6 Sol | 0.441 | LLMs |
| OpenAI GPT-5.5 | 0.44 | LLMs |
| OpenAI GPT-5.6 Terra | 0.44 | LLMs |
| OpenAI GPT-4.1 | 0.438 | LLMs |
Translations from Mistral Large 3, with the most tokens per character, cost $1.55 per million characters, and those from OpenAI GPT-4.1, with the fewest, $4.02. So tokens per character do not set the cost alone. With almost the same tokens per character, translations from the five OpenAI models with a tokenizer result cost $0.63 to $15.93 per million characters. Both figures are lower bounds, from models with reasoning.
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| Microsoft Translator | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.5 | MT engines |
| OpenAI gpt-oss-120b | 3.3 | Open-weight LLMs |
| Translated Lara 2 | 3.6 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 22 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 42 | LLMs |
| DeepSeek V3.2 | 61 | LLMs |
| OpenAI GPT-4.1 | 63 | LLMs |
| OpenAI GPT-5.6 Terra | 70 | LLMs |
| xAI Grok 4.20 (reasoning) | 79 | LLMs |
| Google Gemini 3.5 Flash | 82 | LLMs |
| Anthropic Claude Haiku 4.5 | 85 | LLMs |
| MiniMax M2.5 | 87 | LLMs |
| Mistral Medium 3.1 | 97 | LLMs |
| Google Gemini 3.1 Pro Preview | 117 | LLMs |
| Cohere Command A Translate | 124 | LLMs |
| Anthropic Claude Sonnet 5 | 155 | LLMs |
| OpenAI GPT-5.5 | 156 | LLMs |
| Mistral Large 3 | 161 | LLMs |
| Anthropic Claude Sonnet 4.6 | 172 | LLMs |
| Anthropic Claude Fable 5 | 182 | LLMs |
| Anthropic Claude Opus 4.8 | 210 | LLMs |
| z.ai GLM-5 | 244 | LLMs |
| OpenAI GPT-5.6 Sol | 250 | LLMs |
| Google Gemma 4 | 312 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 434 | 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 Uyghur.
After the three MT engines, the next fastest models are an open-weight LLM and an LLM-based MT engine. A text into Uyghur takes 2.45 seconds with OpenAI gpt-oss-120b and 2.66 seconds with Translated Lara 2, and both count as real-time across the languages we timed. Google Gemma 4, the other open-weight LLM with translations, has the second slowest time of the 27 models.
Leave feedback