The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Maltese 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 Maltese. |
| Tone of voice | The formal register 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 Maltese | 90% band | Minimum | Result |
|---|---|---|---|---|---|
| Similarity | LaBSE, Cohere embed v3 | 0.465 | 0.451 to 0.479 | 0.461 | On the threshold |
| Quality estimation | COMET-Kiwi, MetricX | — | — | 0.534 | Cannot be applied |
We have MetricX scores in Maltese, but for the quality-estimation pair we need COMET-Kiwi scores as well. Besides cost and speed, the page reports the tokens per character of 25 models and the length of a translation into Maltese against the English source. None of them tells how many errors a translation has or which requirements it meets.
Pick a quality level, or your own requirements, bars and constraints, to see the shortlisted solutions that meet them in Maltese, 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 Maltese, so the explorer has none: why.
A time limit of 5 seconds per 1,000 words leaves 10 models in Maltese: the 9 real-time (<2 sec) ones and Yandex Cloud Translate v2. The open-weight OpenAI gpt-oss-120b is the only LLM among them. A limit of 28 seconds per 1,000 words adds Alibaba Machine Translation API and three LLMs: OpenAI GPT-5.6 Luna, Cohere Tiny Aya and xAI Grok 4.20 (non-reasoning). Translations from these three have 0.326 to 0.367 tokens per character, below the median of 0.38.
Translations from Microsoft Translator cost $10 per million characters and take 0.2 seconds per 1,000 words. Those from Yandex Cloud Translate v2 cost $3.94 per million characters and take 3.5 seconds per 1,000 words. Those from OpenAI GPT-5.6 Luna cost $0.48 per million characters, a lower bound, and take 17 seconds per 1,000 words. Translations from Microsoft Translator and Yandex Cloud Translate v2 cost less than those from every faster model with a cost.
Translations from the open-weight Alibaba Qwen3-235B-A22B cost $1.31 per million characters and take 143 seconds per 1,000 words. Those from the open-weight Google Gemma 4 cost less and take longer. We have no price and no tokenizer result for the two other open-weight LLMs, OpenAI gpt-oss-120b and Swiss AI Apertus-8B-2509. With them, a translation takes 3.4 and 53 seconds per 1,000 words.
We cannot assess general translation quality in Maltese: how confident we are.
We cannot assess general translation quality in Maltese: how confident we are.
We cannot assess general translation quality in Maltese: how confident we are.
No translation into Maltese: Tilde TildeOpen LLM engine (far too short).
Of the 38 models we got translations into Maltese from, 8 are MT engines and 3 are LLM-based MT engines: DeepL Next Gen, Google Translation LLM and Translated Lara 2. The others are LLMs, 4 of them open-weight. For each vendor, we got translations from all of its models or from none.
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.16 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | $0.48 | LLMs |
| MiniMax M2.5 | $0.68 | LLMs |
| Mistral Large 3 | $0.84 | LLMs |
| DeepSeek V3.2 | $1.06 | LLMs |
| Mistral Medium 3.1 | $1.06 | LLMs |
| xAI Grok 4.20 (non-reasoning) | $1.27 | LLMs |
| xAI Grok 4.20 (reasoning) | $1.27 | LLMs |
| Alibaba Qwen3-235B-A22B | $1.31 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | $1.42 | LLMs |
| z.ai GLM-5 | $2.13 | LLMs |
| Anthropic Claude Haiku 4.5 | $2.51 | LLMs |
| AI21 Jamba Large 1.7 | $3.02 | LLMs |
| OpenAI GPT-4.1 | $3.26 | LLMs |
| Yandex Cloud Translate v2 | $3.94 | MT engines |
| Google Gemini 3.5 Flash | $4.02 | LLMs |
| OpenAI GPT-5.6 Terra | $4.78 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.42 | LLMs |
| Anthropic Claude Sonnet 5 | $6.46 | LLMs |
| Youdao Translation Service API | $6.71 | MT engines |
| Alibaba Machine Translation API | $7 | MT engines |
| Anthropic Claude Sonnet 4.6 | $7.48 | LLMs |
| OpenAI GPT-5.6 Sol | $8.2 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $11.88 | LLMs |
| Amazon Translate | $15 | MT engines |
| SYSTRAN PNMT API | $15.49 | MT engines |
| Anthropic Claude Opus 4.8 | $16.19 | LLMs |
| Google Cloud Advanced Translation API | $20 | MT engines |
| Google Translation LLM | $21.23 | 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 | $32.49 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Fable 5 | 0.541 | LLMs |
| Anthropic Claude Sonnet 5 | 0.541 | LLMs |
| Anthropic Claude Opus 4.8 | 0.54 | LLMs |
| Mistral Medium 3.1 | 0.452 | LLMs |
| Mistral Large 3 | 0.445 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.419 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.418 | LLMs |
| DeepSeek V3.2 | 0.403 | LLMs |
| MoonshotAI Kimi K2.5 | 0.402 | LLMs |
| Cohere Command A Translate | 0.393 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.392 | Open-weight LLMs |
| z.ai GLM-5 | 0.387 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.38 | LLMs |
| Google Gemini 3.5 Flash | 0.376 | LLMs |
| Google Gemma 4 | 0.373 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | 0.37 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.367 | LLMs |
| MiniMax M2.5 | 0.348 | LLMs |
| OpenAI GPT-5.6 Luna | 0.335 | LLMs |
| OpenAI GPT-5.6 Sol | 0.335 | LLMs |
| OpenAI GPT-5.5 | 0.334 | LLMs |
| OpenAI GPT-4.1 | 0.332 | LLMs |
| OpenAI GPT-5.6 Terra | 0.332 | LLMs |
| Cohere Tiny Aya | 0.326 | LLMs |
| AI21 Jamba Large 1.7 | 0.308 | LLMs |
Over the 25 models with a tokenizer result, translations into Maltese have a median 0.38 tokens per character, 1.6 times as many as English text. Translations from Mistral Large 3 have more tokens per character than those from Anthropic Claude Haiku 4.5, 0.445 against 0.418, and cost less, $0.84 against $2.51 per million characters. So tokens per character alone do not set the cost.
| 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 | 0.6 | LLM-based MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| Amazon Translate | 1.5 | MT engines |
| Translated Lara 2 | 1.6 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 3.4 | Open-weight LLMs |
| SYSTRAN PNMT API | 3.5 | MT engines |
| Yandex Cloud Translate v2 | 3.5 | MT engines |
| OpenAI GPT-5.6 Luna | 17 | LLMs |
| Cohere Tiny Aya | 25 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 26 | LLMs |
| Alibaba Machine Translation API | 28 | MT engines |
| OpenAI GPT-4.1 | 34 | LLMs |
| AI21 Jamba Large 1.7 | 40 | LLMs |
| OpenAI GPT-5.6 Terra | 48 | LLMs |
| Anthropic Claude Haiku 4.5 | 52 | LLMs |
| Swiss AI Apertus-8B-2509 | 53 | Open-weight LLMs |
| Youdao Translation Service API | 56 | MT engines |
| MiniMax M2.5 | 58 | LLMs |
| Mistral Medium 3.1 | 61 | LLMs |
| Cohere Command A Translate | 66 | LLMs |
| xAI Grok 4.20 (reasoning) | 68 | LLMs |
| DeepSeek V3.2 | 69 | LLMs |
| Google Gemini 3.5 Flash | 73 | LLMs |
| Anthropic Claude Sonnet 5 | 84 | LLMs |
| OpenAI GPT-5.5 | 85 | LLMs |
| Anthropic Claude Fable 5 | 89 | LLMs |
| Anthropic Claude Sonnet 4.6 | 96 | LLMs |
| Google Gemini 3.1 Pro Preview | 96 | LLMs |
| Mistral Large 3 | 107 | LLMs |
| Anthropic Claude Opus 4.8 | 113 | LLMs |
| OpenAI GPT-5.6 Sol | 129 | LLMs |
| Alibaba Qwen3-235B-A22B | 143 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 234 | LLMs |
| z.ai GLM-5 | 251 | LLMs |
| Google Gemma 4 | 337 | Open-weight 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 Maltese.
With the three MT engines that are not real-time, a translation into Maltese takes 3.5 to 56 seconds per 1,000 words. A text takes 2.51 seconds with the open-weight OpenAI gpt-oss-120b and 2.63 seconds with SYSTRAN PNMT API, real-time across the languages we timed, and 2.6 seconds with Yandex Cloud Translate v2, which is not. Translations from xAI Grok 4.20 (reasoning) take 68 seconds per 1,000 words, and those from xAI Grok 4.20 (non-reasoning) take 26 seconds per 1,000 words.
Of the alternatives to neutral rephrasing in Maltese, a masculine default does not meet gender neutrality, because the requirement does not allow a default gender. With no shortlist, this page does not report a tone of voice or gender neutrality result in Maltese.
Leave feedback