The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Malay 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 Malay. |
| 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 Malay, 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. Inconclusive: the three LQA judges agree on fewer than 90% of the sentences of the solution, so it has no general translation quality result (09-02). 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.
Three one-step solutions meet every Standard requirement in Malay, and none leads on errors, cost and time at once. The one on OpenAI GPT-5.5 has the fewest total errors, 10.9 per 1,000 words, but costs $9.82 per million characters and takes 50 seconds per 1,000 words. The one on OpenAI GPT-5.6 Luna, with 12.2 total errors per 1,000 words, is the cheapest and ties as the fastest, at 13 seconds per 1,000 words, with the one on Google Gemini 3.5 Flash, which has more errors than either.
Among these three, a cost cap or a time limit under those of the solution on OpenAI GPT-5.5, $9.82 per million characters and 50 seconds per 1,000 words, removes it. A cap under $2.92 per million characters also removes the one on Google Gemini 3.5 Flash. A cap or a limit under those of the one on OpenAI GPT-5.6 Luna, $0.39 per million characters and 13 seconds per 1,000 words, removes all three. All three take one step, so a limit on the steps removes none.
Translations from the two best solutions in Malay have 100% of the sentences about people gender-neutral without the gender-neutrality agent. So the Advanced requirements add nothing to their cost, time or total errors. The cheapest solution that meets every Advanced requirement is the one on OpenAI GPT-5.6 Luna, at $0.39 per million characters, as for the Standard requirements, and it is also the fastest.
OpenAI GPT-5.5
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
Tied with Google Gemini 3.5 Flash (one-step LLM adjusted for Standard requirements).
OpenAI GPT-5.5
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
Tied with Google Gemini 3.5 Flash (one-step LLM adjusted for Standard requirements).
OpenAI GPT-5.5
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 | ★★★ | one-step | OpenAI GPT-5.6 Luna | 12.2 | $0.39 | 13 |
| Standard | Tier 1: simple solution + backup | ★★★ | one-step | OpenAI GPT-5.6 Luna | 12.2 | $0.39 | 13 |
| Advanced | Tier 1: simple solution + backup | ★★ | one-step | OpenAI GPT-5.6 Luna | 12.2 | $0.39 | 13 |
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 three solutions that meet every Basic requirement in Malay also meet every Standard requirement. Two of them, on OpenAI GPT-5.5 and OpenAI GPT-5.6 Luna, meet every Advanced requirement too. Each level takes a single step. At all three levels, Optimize for selects the solution on OpenAI GPT-5.5 for quality and the one on OpenAI GPT-5.6 Luna for cost and latency, where under the Basic and Standard requirements it ties with the one on Google Gemini 3.5 Flash at 13 seconds per 1,000 words.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-5.6 Luna | 0.39 | 12.2 | Meets every Standard requirement |
| Mistral Large 3 | 0.8 | 36.3 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.77 | 21.4 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 2.92 | 17.5 | Meets every Standard requirement |
| Anthropic Claude Sonnet 4.6 | 6.68 | 26.4 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 9.82 | 10.9 | Meets every Standard requirement |
| DeepL Next Gen | 27.5 | 40.9 | Misses a Standard requirement |
A higher price does not on its own mean fewer errors among the one-step solutions in Malay. The cost and errors front holds two LLM solutions, on OpenAI GPT-5.6 Luna and OpenAI GPT-5.5. Among the one-step solutions with a cost, the one on Mistral Large 3 is the second cheapest, at $0.80 per million characters. It has 36.3 weighted errors per 1,000 words. The one on DeepL Next Gen costs $27.50 per million characters and has 40.9 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 26.4 |
| 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.
General translation quality holds back most of the shortlist in Malay. Over its bar, the solution on OpenAI GPT-4.1 has 21.4 weighted errors per 1,000 words and the one on Anthropic Claude Sonnet 4.6 has 26.4. The other four solutions over the bar have 36.3 to 43.5 weighted errors per 1,000 words. A solution meets tag handling with 100% of the tags right. Translations from OpenAI gpt-oss-120b have 99.6% of the tags right, from Anthropic Claude Sonnet 4.6 99.2%, from DeepL Next Gen 95.7% and from AI21 Jamba Large 1.7 92.3%.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Malay | Class |
|---|---|---|
| OpenAI gpt-oss-120b | 43.5 | Open-weight LLMs |
| Cohere Command A Translate | 42.2 | LLMs |
| DeepL Next Gen | 40.9 | LLM-based MT engines |
| Mistral Large 3 | 36.3 | LLMs |
| Anthropic Claude Sonnet 4.6 | 26.4 | LLMs |
| OpenAI GPT-4.1 | 21.4 | LLMs |
| Google Gemini 3.5 Flash | 17.5 | LLMs |
| OpenAI GPT-5.6 Luna | 12.2 | LLMs |
| OpenAI GPT-5.5 | 10.9 | LLMs |
No translation into Malay: Widn AI Tower Anthill (far too short, wrong language), Widn AI Tower Vesuvius (far too short, wrong language).
Model type alone does not order the results in Malay. All three solutions that meet general translation quality are on LLMs, but translations from the eight shortlisted LLMs with a conclusive result span 10.9 to 43.5 weighted errors per 1,000 words, and those from DeepL Next Gen, with 40.9, fall inside that span. The only MT engines on the shortlist are LLM-based: DeepL Next Gen and Translated Lara 2. We got no translation from Widn AI Tower Anthill and Widn AI Tower Vesuvius: their output was far too short and in the wrong language.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | 0.787 | 9 | 1.49 | 10 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.787 | 10 | 1.53 | 14 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.787 | 11 | 1.52 | 13 | No | Not shortlisted |
| z.ai GLM-5 | 0.787 | 12 | 1.56 | 15 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.786 | 13 | 1.59 | 18 | No | Not shortlisted |
| Google Gemma 4 | 0.785 | 16 | 1.47 | 9 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.784 | 17 | 1.71 | 25 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.779 | 18 | 1.67 | 24 | No | Not shortlisted |
| Google Translation LLM | 0.778 | 20 | 1.81 | 26 | No | Not shortlisted |
| DeepSeek V3.2 | 0.778 | 21 | 1.57 | 17 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.778 | 22 | 1.62 | 22 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.777 | 23 | 1.60 | 20 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.776 | 24 | 1.56 | 16 | No | Not shortlisted |
| Microsoft Translator | 0.769 | 25 | 1.60 | 19 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.762 | 26 | 1.60 | 21 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.751 | 27 | 2.36 | 29 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.745 | 28 | 2.84 | 33 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.745 | 29 | 2.08 | 27 | No | Not shortlisted |
| Tencent Cloud TMT | 0.741 | 30 | 2.28 | 28 | No | Not shortlisted |
| MiniMax M2.5 | 0.739 | 31 | 2.45 | 31 | No | Not shortlisted |
| Amazon Translate | 0.737 | 32 | 2.37 | 30 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.696 | 33 | 2.66 | 32 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.688 | 34 | 3.29 | 34 | No | Not shortlisted |
| AppTek NMT | 0.657 | 35 | 3.90 | 35 | No | Not shortlisted |
| Cohere Tiny Aya | 0.653 | 36 | 5.09 | 36 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.794 | 1 | 1.35 | 1 | Yes | Top 8 by both metrics |
| Cohere Command A Translate | 0.792 | 2 | 1.52 | 12 | Yes | Top 8 by COMET-Kiwi |
| DeepL Next Gen | 0.791 | 3 | 1.41 | 4 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.791 | 4 | 1.38 | 2 | Yes | Top 8 by both metrics |
| OpenAI gpt-oss-120b | 0.790 | 5 | 1.47 | 8 | Yes | Top 8 by both metrics |
| Mistral Large 3 | 0.789 | 6 | 1.42 | 5 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.788 | 7 | 1.38 | 3 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.787 | 8 | 1.51 | 11 | Yes | Top 8 by COMET-Kiwi |
| AI21 Jamba Large 1.7 | 0.786 | 14 | 1.44 | 6 | Yes | Top 8 by MetricX |
| Anthropic Claude Sonnet 4.6 | 0.785 | 15 | 1.47 | 7 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.779 | 19 | 1.63 | 23 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-4.1 | Adjusted LLM | 1 | 0.794 | 1 | 1.35 | Top 8 by both metrics |
| Cohere Command A Translate | Adjusted LLM | 2 | 0.792 | 12 | 1.52 | Top 8 by COMET-Kiwi |
| DeepL Next Gen | Off-the-shelf NMT | 3 | 0.791 | 4 | 1.41 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 4 | 0.791 | 2 | 1.38 | Top 8 by both metrics |
| OpenAI gpt-oss-120b | Adjusted LLM | 5 | 0.790 | 8 | 1.47 | Top 8 by both metrics |
| Mistral Large 3 | Adjusted LLM | 6 | 0.789 | 5 | 1.42 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 7 | 0.788 | 3 | 1.38 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 8 | 0.787 | 11 | 1.51 | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 9 | 0.787 | 10 | 1.49 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 10 | 0.787 | 14 | 1.53 | Not shortlisted |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 11 | 0.787 | 13 | 1.52 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 12 | 0.787 | 15 | 1.56 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 13 | 0.786 | 18 | 1.59 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 14 | 0.786 | 6 | 1.44 | Top 8 by MetricX |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 15 | 0.785 | 7 | 1.47 | Top 8 by MetricX |
| Google Gemma 4 | Adjusted LLM | 16 | 0.785 | 9 | 1.47 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 17 | 0.784 | 25 | 1.71 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 18 | 0.779 | 24 | 1.67 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 19 | 0.779 | 23 | 1.63 | Real-time place |
| Google Translation LLM | Adjusted NMT | 20 | 0.778 | 26 | 1.81 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 21 | 0.778 | 17 | 1.57 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 22 | 0.778 | 22 | 1.62 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.777 | 20 | 1.60 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 24 | 0.776 | 16 | 1.56 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 25 | 0.769 | 19 | 1.60 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 26 | 0.762 | 21 | 1.60 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 27 | 0.751 | 29 | 2.36 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 28 | 0.745 | 33 | 2.84 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 29 | 0.745 | 27 | 2.08 | Not shortlisted |
| Tencent Cloud TMT | Off-the-shelf NMT | 30 | 0.741 | 28 | 2.28 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 31 | 0.739 | 31 | 2.45 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 32 | 0.737 | 30 | 2.37 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 33 | 0.696 | 32 | 2.66 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 34 | 0.688 | 34 | 3.29 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 35 | 0.657 | 35 | 3.90 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 36 | 0.653 | 36 | 5.09 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.788 | — | 1.40 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.785 | — | 1.51 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.771 | — | 1.77 | 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 36 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 metrics and Intento LQA order the shortlist differently in Malay. The solution on OpenAI GPT-4.1, first on both metrics, misses general translation quality, while the one on OpenAI GPT-5.6 Luna, 7th on COMET-Kiwi and 3rd on MetricX, meets every Advanced requirement. COMET-Kiwi alone brought in Cohere Command A Translate and OpenAI GPT-5.5, and MetricX alone AI21 Jamba Large 1.7 and Anthropic Claude Sonnet 4.6. Nearest the cut are Google Gemini 3.1 Pro Preview, 9th on COMET-Kiwi, and Google Gemma 4, 9th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | $0.39 |
| Standard | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | $0.39 |
| Advanced | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | $0.39 |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.4 | LLMs |
| Mistral Large 3 | $0.8 | LLMs |
| OpenAI GPT-4.1 | $2.77 | LLMs |
| Google Gemini 3.5 Flash | $2.92 | LLMs |
| AI21 Jamba Large 1.7 | $3.62 | LLMs |
| Anthropic Claude Sonnet 4.6 | $6.68 | LLMs |
| OpenAI GPT-5.5 | $9.82 | 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 Fable 5 | 0.479 | LLMs |
| Anthropic Claude Sonnet 5 | 0.479 | LLMs |
| Anthropic Claude Opus 4.8 | 0.478 | LLMs |
| Mistral Medium 3.1 | 0.403 | LLMs |
| Mistral Large 3 | 0.399 | LLMs |
| AI21 Jamba Large 1.7 | 0.374 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.354 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.353 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.308 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 0.307 | LLMs |
| z.ai GLM-5 | 0.302 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.283 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.283 | LLMs |
| DeepSeek V3.2 | 0.281 | LLMs |
| Cohere Command A Translate | 0.263 | LLMs |
| OpenAI GPT-4.1 | 0.261 | LLMs |
| OpenAI GPT-5.6 Terra | 0.261 | LLMs |
| OpenAI GPT-5.5 | 0.26 | LLMs |
| OpenAI GPT-5.6 Luna | 0.259 | LLMs |
| OpenAI GPT-5.6 Sol | 0.259 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.253 | LLMs |
| Google Gemini 3.5 Flash | 0.253 | LLMs |
| Google Gemma 4 | 0.251 | Open-weight LLMs |
| MiniMax M2.5 | 0.225 | LLMs |
| Cohere Tiny Aya | 0.217 | LLMs |
The tokenizers produce a median 0.28 tokens per character of a Malay translation, 1.2 times their count for English. The three models whose solutions meet every Standard requirement have tokenizers near the low end, at 0.25 to 0.26 tokens per character. Yet their solutions cost from $0.39 to $9.82 per million characters. No level takes more than one step in Malay, so no step adds to the cost, and the cheapest solution is the same at all three levels.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | 13 |
| Standard | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | 13 |
| Advanced | OpenAI GPT-5.6 Luna, one-step LLM adjusted for Standard requirements | 13 |
| 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 | 0.3 | MT engines |
| Google Translation LLM | 0.4 | LLM-based MT engines |
| DeepL Next Gen | 0.5 | LLM-based MT engines |
| SYSTRAN PNMT API | 0.9 | MT engines |
| Translated Lara 2 | 1.2 | LLM-based MT engines |
| Amazon Translate | 1.2 | MT engines |
| OpenAI gpt-oss-120b | 1.9 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 2.3 | MT engines |
| Google Gemini 3.5 Flash | 13 | LLMs |
| OpenAI GPT-5.6 Luna | 13 | LLMs |
| Cohere Tiny Aya | 13 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 13 | LLMs |
| Alibaba Machine Translation API | 19 | MT engines |
| OpenAI GPT-4.1 | 21 | LLMs |
| OpenAI GPT-5.6 Terra | 22 | LLMs |
| Anthropic Claude Haiku 4.5 | 30 | LLMs |
| DeepSeek V3.2 | 31 | LLMs |
| Cohere Command A Translate | 34 | LLMs |
| Mistral Medium 3.1 | 36 | LLMs |
| MiniMax M2.5 | 37 | LLMs |
| Swiss AI Apertus-8B-2509 | 38 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | 40 | LLMs |
| Anthropic Claude Sonnet 5 | 47 | LLMs |
| OpenAI GPT-5.6 Sol | 49 | LLMs |
| OpenAI GPT-5.5 | 50 | LLMs |
| AI21 Jamba Large 1.7 | 50 | LLMs |
| Anthropic Claude Fable 5 | 51 | LLMs |
| Mistral Large 3 | 52 | LLMs |
| Tencent Cloud TMT | 55 | MT engines |
| Anthropic Claude Opus 4.8 | 60 | LLMs |
| Anthropic Claude Sonnet 4.6 | 60 | LLMs |
| Google Gemini 3.1 Pro Preview | 76 | LLMs |
| AppTek NMT | 93 | MT engines |
| Alibaba Qwen3-235B-A22B | 108 | Open-weight LLMs |
| Google Gemma 4 | 122 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 215 | LLMs |
| z.ai GLM-5 | 590 | 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 Malay.
| 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-5.5 | one-step | 100% | ✓ | $9.82 | 50 |
| OpenAI GPT-5.6 Luna | one-step | 100% | ✓ | $0.39 | 13 |
In Malay we check tone of voice against the formal register, with “anda” as the form of address. For gender neutrality, a translation must not add a gender the source does not specify, and a default gender is never allowed. Malay has no grammatical gender for persons, so translations are neutral by default, as long as they have no gendered terms. Translations from all 11 shortlisted one-step solutions have 100% of their sentences in the tone of voice.
Leave feedback