The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Indonesian 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 Indonesian. |
| Tone of voice | The formal register (Anda) 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 Indonesian, 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 12 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.
Optimize for selects different solutions for each goal among the six that meet every Standard requirement. For quality it selects Google Gemini 3.1 Pro Preview, with 12.5 total errors per 1,000 words, at $3.74 per million characters and 76 seconds per 1,000 words. For cost it selects Google Gemma 4, at $0.11 per million characters and 115 seconds per 1,000 words, with 17.8 total errors per 1,000 words. For latency Google Gemini 3.5 Flash and OpenAI GPT-5.6 Luna tie at 12 seconds per 1,000 words.
Of the six solutions that meet every Standard requirement, a cost cap from $0.38 to $1.76 per million characters leaves those on Google Gemma 4 and OpenAI GPT-5.6 Luna. Among the same six, a time limit of 20 seconds per 1,000 words leaves those on Google Gemini 3.5 Flash and OpenAI GPT-5.6 Luna. Together they leave the one on OpenAI GPT-5.6 Luna alone, with 14.2 total errors per 1,000 words, second only to Google Gemini 3.1 Pro Preview. All six are one-step solutions, so a limit on the steps keeps all of them.
The Advanced requirements leave two of the six, the two best solutions, on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.6 Luna. Translations from both have 100% of the sentences about people gender-neutral without the gender-neutrality agent, so their cost and time stay the same. The cheapest solution changes from Google Gemma 4 at $0.11 to OpenAI GPT-5.6 Luna at $0.38 per million characters, and the fastest is OpenAI GPT-5.6 Luna alone, at 12 seconds per 1,000 words, no longer tied with Google Gemini 3.5 Flash.
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemma 4
one-step LLM adjusted for Standard requirements
Google Gemini 3.5 Flash
one-step LLM adjusted for Standard requirements
Tied with OpenAI GPT-5.6 Luna (one-step LLM adjusted for Standard requirements).
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemma 4
one-step LLM adjusted for Standard requirements
Google Gemini 3.5 Flash
one-step LLM adjusted for Standard requirements
Tied with OpenAI GPT-5.6 Luna (one-step LLM adjusted for Standard requirements).
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Luna
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Luna
one-step LLM adjusted for Standard requirements
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 | ★×7 | one-step | Google Gemma 4 | 17.8 | $0.11 | 115 |
| Standard | Tier 1: simple solution + backup | ★×6 | one-step | Google Gemma 4 | 17.8 | $0.11 | 115 |
| Advanced | Tier 1: simple solution + backup | ★★ | one-step | OpenAI GPT-5.6 Luna | 14.2 | $0.38 | 12 |
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.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.11 | 17.8 | Meets every Standard requirement |
| OpenAI GPT-5.6 Luna | 0.38 | 14.2 | Meets every Standard requirement |
| Mistral Medium 3.1 | 0.99 | 56 | Misses a Standard requirement |
| z.ai GLM-5 | 1.77 | 15.5 | Meets every Standard requirement |
| OpenAI GPT-4.1 | 2.69 | 18.1 | Meets every Standard requirement |
| Google Gemini 3.5 Flash | 2.79 | 17.5 | Meets every Standard requirement |
| Google Gemini 3.1 Pro Preview | 3.74 | 12.5 | Meets every Standard requirement |
| Anthropic Claude Sonnet 5 | 5.98 | 19.8 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 33 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 21.1 | Misses a Standard requirement |
The front of cost against errors holds three LLM solutions: the open-weight Google Gemma 4 at $0.11, OpenAI GPT-5.6 Luna at $0.38 and Google Gemini 3.1 Pro Preview at $3.74 per million characters. Every other solution with a price costs more than the one on OpenAI GPT-5.6 Luna and has more weighted errors. The two dearest of them, on DeepL Next Gen at $27.50 and Translated Lara 2 at $24.99 per million characters, have 21.1 and 33.0 weighted errors per 1,000 words, against 17.8 from Google Gemma 4.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 19 |
| Tone of voice | 0 |
| Tag handling | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Translations from DeepL Next Gen miss general translation quality with 21.1 weighted errors per 1,000 words, just over the bar, no critical error and three major errors. Tag handling asks for every tag, and translations from Cohere Command A Translate keep 99.6% of the tags, from Mistral Medium 3.1 97.6%, from Anthropic Claude Sonnet 5 96.1% and from DeepL Next Gen 95.7%. The solutions on both LLM-based MT engines, DeepL Next Gen and Translated Lara 2, miss formatting, and so does the one on Mistral Medium 3.1, an LLM.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Indonesian | Class |
|---|---|---|
| Mistral Medium 3.1 | 56 | LLMs |
| OpenAI gpt-oss-120b | 33.6 | Open-weight LLMs |
| Translated Lara 2 | 33 | LLM-based MT engines |
| Cohere Command A Translate | 32.3 | LLMs |
| DeepL Next Gen | 21.1 | LLM-based MT engines |
| Anthropic Claude Sonnet 5 | 19.8 | LLMs |
| OpenAI GPT-4.1 | 18.1 | LLMs |
| Google Gemma 4 | 17.8 | Open-weight LLMs |
| Google Gemini 3.5 Flash | 17.5 | LLMs |
| z.ai GLM-5 | 15.5 | LLMs |
| OpenAI GPT-5.6 Luna | 14.2 | LLMs |
| Google Gemini 3.1 Pro Preview | 12.5 | LLMs |
No translation into Indonesian: Widn AI Tower Anthill (far too short, wrong language), Widn AI Tower Vesuvius (far too short, wrong language).
Translations from the eight LLM solutions on the shortlist have 12.5 to 56.0 weighted errors per 1,000 words, from the two open-weight LLMs, Google Gemma 4 and OpenAI gpt-oss-120b, 17.8 and 33.6, and from the two LLM-based MT engines 21.1 and 33.0. None of the MT engines that are not LLM-based made the shortlist. The highest, 56.0 weighted errors per 1,000 words, is from Mistral Medium 3.1, the shortlisted LLM that does not meet our language-support criteria, with three critical errors.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Opus 4.8 | 0.809 | 10 | 1.25 | 12 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.809 | 11 | 1.26 | 13 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.808 | 12 | 1.24 | 9 | No | Not shortlisted |
| OpenAI GPT-5.5 | 0.808 | 13 | 1.30 | 19 | No | Not shortlisted |
| Mistral Large 3 | 0.808 | 14 | 1.29 | 16 | No | Not shortlisted |
| DeepSeek V3.2 | 0.804 | 16 | 1.32 | 20 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.803 | 17 | 1.27 | 15 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.803 | 18 | 1.34 | 21 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.802 | 19 | 1.35 | 22 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.799 | 22 | 1.39 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.797 | 23 | 1.26 | 14 | No | Not shortlisted |
| Microsoft Translator | 0.797 | 24 | 1.30 | 18 | No | Not shortlisted |
| Meta Llama 4 Maverick | 0.796 | 25 | 1.38 | 24 | No | Not shortlisted |
| Google Translation LLM | 0.795 | 26 | 1.95 | 31 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.790 | 27 | 1.46 | 26 | No | Not shortlisted |
| Amazon Translate | 0.785 | 28 | 1.49 | 27 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.784 | 29 | 1.54 | 28 | No | Not shortlisted |
| MiniMax M2.5 | 0.784 | 30 | 1.67 | 29 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.780 | 31 | 2.11 | 33 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.777 | 32 | 1.77 | 30 | No | Not shortlisted |
| Naver Papago NMT Commercial | 0.752 | 33 | 2.00 | 32 | No | Not shortlisted |
| Tencent Cloud TMT | 0.747 | 34 | 2.40 | 35 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.735 | 35 | 2.14 | 34 | No | Not shortlisted |
| AppTek NMT | 0.684 | 36 | 3.18 | 36 | No | Not shortlisted |
| Cohere Tiny Aya | 0.669 | 37 | 4.52 | 38 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.639 | 38 | 3.92 | 37 | No | Not shortlisted |
| DeepL Next Gen | 0.817 | 1 | 1.13 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.812 | 2 | 1.16 | 2 | Yes | Top 8 by both metrics |
| Cohere Command A Translate | 0.811 | 3 | 1.21 | 5 | Yes | Top 8 by both metrics |
| Google Gemma 4 | 0.811 | 4 | 1.22 | 6 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.810 | 5 | 1.25 | 11 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | 0.810 | 6 | 1.23 | 7 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.810 | 7 | 1.23 | 8 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.810 | 8 | 1.25 | 10 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Luna | 0.809 | 9 | 1.20 | 4 | Yes | Top 8 by MetricX |
| OpenAI gpt-oss-120b | 0.806 | 15 | 1.29 | 17 | Yes | Real-time place |
| Mistral Medium 3.1 | 0.802 | 20 | 1.20 | 3 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.801 | 21 | 1.37 | 23 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Off-the-shelf NMT | 1 | 0.817 | 1 | 1.13 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 2 | 0.812 | 2 | 1.16 | Top 8 by both metrics |
| Cohere Command A Translate | Adjusted LLM | 3 | 0.811 | 5 | 1.21 | Top 8 by both metrics |
| Google Gemma 4 | Adjusted LLM | 4 | 0.811 | 6 | 1.22 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 5 | 0.810 | 11 | 1.25 | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 6 | 0.810 | 7 | 1.23 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 7 | 0.810 | 8 | 1.23 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 8 | 0.810 | 10 | 1.25 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 9 | 0.809 | 4 | 1.20 | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 10 | 0.809 | 12 | 1.25 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 11 | 0.809 | 13 | 1.26 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 12 | 0.808 | 9 | 1.24 | Not shortlisted |
| OpenAI GPT-5.5 | Adjusted LLM | 13 | 0.808 | 19 | 1.30 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 14 | 0.808 | 16 | 1.29 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 15 | 0.806 | 17 | 1.29 | Real-time place |
| DeepSeek V3.2 | Adjusted LLM | 16 | 0.804 | 20 | 1.32 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 17 | 0.803 | 15 | 1.27 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 18 | 0.803 | 21 | 1.34 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 19 | 0.802 | 22 | 1.35 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 20 | 0.802 | 3 | 1.20 | Top 8 by MetricX |
| Translated Lara 2 | Adjusted NMT | 21 | 0.801 | 23 | 1.37 | Real-time place |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 22 | 0.799 | 25 | 1.39 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 23 | 0.797 | 14 | 1.26 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 24 | 0.797 | 18 | 1.30 | Not shortlisted |
| Meta Llama 4 Maverick | Adjusted LLM | 25 | 0.796 | 24 | 1.38 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 26 | 0.795 | 31 | 1.95 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 27 | 0.790 | 26 | 1.46 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 28 | 0.785 | 27 | 1.49 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 29 | 0.784 | 28 | 1.54 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 30 | 0.784 | 29 | 1.67 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 31 | 0.780 | 33 | 2.11 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 32 | 0.777 | 30 | 1.77 | Not shortlisted |
| Naver Papago NMT Commercial | Off-the-shelf NMT | 33 | 0.752 | 32 | 2.00 | Not shortlisted |
| Tencent Cloud TMT | Off-the-shelf NMT | 34 | 0.747 | 35 | 2.40 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 35 | 0.735 | 34 | 2.14 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 36 | 0.684 | 36 | 3.18 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 37 | 0.669 | 38 | 4.52 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 38 | 0.639 | 37 | 3.92 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.803 | — | 1.24 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.809 | — | 1.25 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.800 | — | 1.36 | 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 Intento LQA order differs from the metric order. The solution on DeepL Next Gen, first on both metrics, misses general translation quality, while translations from Google Gemini 3.1 Pro Preview, sixth on COMET-Kiwi and seventh on MetricX, have the fewest weighted errors, 12.5 per 1,000 words. MetricX alone brought in OpenAI GPT-5.6 Luna, ninth on COMET-Kiwi, and its solution meets every Advanced requirement. Nearest the cut are Anthropic Claude Opus 4.8 on COMET-Kiwi and Anthropic Claude Sonnet 4.6, ninth on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemma 4, one-step LLM adjusted for Standard requirements | $0.11 |
| Standard | Google Gemma 4, one-step LLM adjusted for Standard requirements | $0.11 |
| Advanced | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | $0.38 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.11 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | $0.38 | LLMs |
| Mistral Medium 3.1 | $0.99 | LLMs |
| z.ai GLM-5 | $1.77 | LLMs |
| OpenAI GPT-4.1 | $2.69 | LLMs |
| Google Gemini 3.5 Flash | $2.79 | LLMs |
| Google Gemini 3.1 Pro Preview | $3.74 | LLMs |
| Anthropic Claude Sonnet 5 | $5.98 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.48 | LLMs |
| Anthropic Claude Sonnet 5 | 0.479 | LLMs |
| Anthropic Claude Fable 5 | 0.478 | LLMs |
| Mistral Medium 3.1 | 0.396 | LLMs |
| Mistral Large 3 | 0.392 | LLMs |
| AI21 Jamba Large 1.7 | 0.369 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.353 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.351 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.303 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 0.303 | LLMs |
| z.ai GLM-5 | 0.294 | LLMs |
| DeepSeek V3.2 | 0.273 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.272 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.272 | LLMs |
| OpenAI GPT-4.1 | 0.252 | LLMs |
| OpenAI GPT-5.5 | 0.252 | LLMs |
| OpenAI GPT-5.6 Terra | 0.252 | LLMs |
| OpenAI GPT-5.6 Luna | 0.251 | LLMs |
| OpenAI GPT-5.6 Sol | 0.25 | LLMs |
| Cohere Command A Translate | 0.245 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.242 | LLMs |
| Google Gemini 3.5 Flash | 0.242 | LLMs |
| Google Gemma 4 | 0.241 | Open-weight LLMs |
| Cohere Tiny Aya | 0.214 | LLMs |
| MiniMax M2.5 | 0.208 | LLMs |
The tokenizers of 25 LLMs produce a median 0.27 tokens per character of Indonesian, 1.2 times their count for English. Five of the six solutions that meet every Standard requirement are on models whose tokenizers produce 0.241 to 0.252 tokens per character, and the sixth, on z.ai GLM-5, 0.294. Every level takes one step, and the cheapest solution is on Google Gemma 4 at $0.11 per million characters under the Basic and Standard requirements and on OpenAI GPT-5.6 Luna at $0.38 under the Advanced ones.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Google Gemini 3.5 Flash, one-step LLM adjusted for Standard requirements | 12 |
| Standard | Google Gemini 3.5 Flash, one-step LLM adjusted for Standard requirements | 12 |
| Advanced | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | 12 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| Google Translation LLM | 0.4 | LLM-based MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Translated Lara 2 | 1.1 | LLM-based MT engines |
| SYSTRAN PNMT API | 1.1 | MT engines |
| Amazon Translate | 1.2 | MT engines |
| OpenAI gpt-oss-120b | 1.9 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.5 | MT engines |
| Cohere Tiny Aya | 11 | LLMs |
| Google Gemini 3.5 Flash | 12 | LLMs |
| OpenAI GPT-5.6 Luna | 12 | LLMs |
| OpenAI GPT-5.6 Terra | 19 | LLMs |
| Alibaba Machine Translation API | 21 | MT engines |
| OpenAI GPT-4.1 | 22 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 22 | LLMs |
| Cohere Command A Translate | 28 | LLMs |
| Anthropic Claude Haiku 4.5 | 28 | LLMs |
| MiniMax M2.5 | 28 | LLMs |
| OpenAI GPT-5.5 | 36 | LLMs |
| Anthropic Claude Sonnet 5 | 37 | LLMs |
| Swiss AI Apertus-8B-2509 | 38 | Open-weight LLMs |
| DeepSeek V3.2 | 39 | LLMs |
| Meta Llama 4 Maverick | 40 | Open-weight LLMs |
| Anthropic Claude Fable 5 | 44 | LLMs |
| OpenAI GPT-5.6 Sol | 46 | LLMs |
| Mistral Large 3 | 47 | LLMs |
| Anthropic Claude Opus 4.8 | 48 | LLMs |
| AI21 Jamba Large 1.7 | 51 | LLMs |
| Tencent Cloud TMT | 53 | MT engines |
| Anthropic Claude Sonnet 4.6 | 55 | LLMs |
| Mistral Medium 3.1 | 65 | LLMs |
| xAI Grok 4.20 (reasoning) | 66 | LLMs |
| Google Gemini 3.1 Pro Preview | 76 | LLMs |
| AppTek NMT | 84 | MT engines |
| Alibaba Qwen3-235B-A22B | 94 | Open-weight LLMs |
| Naver Papago NMT Commercial | 108 | MT engines |
| Google Gemma 4 | 115 | Open-weight LLMs |
| z.ai GLM-5 | 184 | LLMs |
| MoonshotAI Kimi K2.5 | 243 | 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 Indonesian.
Five of the nine real-time models are MT engines, three LLM-based MT engines and one the open-weight OpenAI gpt-oss-120b. None of the three on the shortlist meets general translation quality: translations from DeepL Next Gen, Translated Lara 2 and OpenAI gpt-oss-120b have 21.1, 33.0 and 33.6 weighted errors per 1,000 words. The two fastest that meet every Standard requirement are on LLMs, Google Gemini 3.5 Flash and OpenAI GPT-5.6 Luna, tied at 12 seconds per 1,000 words, and only the second meets every Advanced requirement.
| 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 |
|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | one-step | 100% | ✓ | $3.74 | 76 |
| OpenAI GPT-5.6 Luna | one-step | 100% | ✓ | $0.38 | 12 |
Tone of voice in Indonesian asks for the formal register, with Anda as the form of address. For gender neutrality, Indonesian has no grammatical gender for persons, so a translation is neutral by default and only has to avoid gendered terms, and the requirement rules out a default gender. Translations from both of the two best solutions have 100% of the sentences about people gender-neutral without the gender-neutrality agent.
The agreement of both pairs of automatic metrics is above the threshold even at the low end of its range, and neither pair is borderline. The three LQA judges agree on a median of 95.5% of the sentences of the shortlisted solutions. So the shortlist, ranked on the metrics, and the Intento LQA results on this page carry their full weight.
Leave feedback