The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Turkish 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 Turkish. |
| Tone of voice | The formal register (siz) 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 Turkish, 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.
Three one-step solutions meet every Standard requirement in Turkish. The one on Google Gemini 3.1 Pro Preview costs $4.23 per million characters, has 18.8 total errors per 1,000 words and takes 98 seconds per 1,000 words. The ones on OpenAI GPT-5.5 and Anthropic Claude Fable 5 have 12.2 total errors per 1,000 words, cost $10.44 and $31.37 per million characters and take 45 and 57 seconds per 1,000 words. Optimize for marks both for quality, as they tie, and selects Google Gemini 3.1 Pro Preview for cost and OpenAI GPT-5.5 for latency.
Under the Standard requirements, a cost cap below $10.44 per million characters leaves at most the solution on Google Gemini 3.1 Pro Preview. It costs $4.23 per million characters and takes 98 seconds per 1,000 words. None of the three meet such a cap together with a time limit shorter than that. A time limit below 57 seconds per 1,000 words leaves at most the solution on OpenAI GPT-5.5, which takes 45 seconds. A limit to one step leaves all three, as each is a one-step solution.
The Advanced requirements leave two of the three solutions, on OpenAI GPT-5.5 and Anthropic Claude Fable 5, and both meet gender neutrality without the gender-neutrality agent. The lowest cost rises from $4.23 to $10.44 per million characters, as the solution on Google Gemini 3.1 Pro Preview drops out. The shortest time is still 45 seconds per 1,000 words, and the fewest total errors are still 12.2 per 1,000 words.
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Tied with OpenAI GPT-5.5 (one-step LLM adjusted for Standard requirements).
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Tied with OpenAI GPT-5.5 (one-step LLM adjusted for Standard requirements).
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Tied with OpenAI GPT-5.5 (one-step LLM adjusted for Standard requirements).
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
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 | ★×5 | one-step | Google Gemini 3.1 Pro Preview | 18.8 | $4.23 | 98 |
| Standard | Tier 1: simple solution + backup | ★★★ | one-step | Google Gemini 3.1 Pro Preview | 18.8 | $4.23 | 98 |
| Advanced | Tier 1: simple solution + backup | ★★ | one-step | OpenAI GPT-5.5 | 12.2 | $10.44 | 45 |
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 | 0.13 | 28.3 | Misses a Standard requirement |
| xAI Grok 4.20 (non-reasoning) | 1.03 | 30.3 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.92 | 29.7 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.14 | 31.3 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.23 | 18.8 | Meets every Standard requirement |
| OpenAI GPT-5.6 Sol | 7.23 | 15.8 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 7.54 | 49.1 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 10.44 | 12.2 | Meets every Standard requirement |
| Anthropic Claude Opus 4.8 | 15.65 | 15.8 | Misses a Standard requirement |
| Google Translation LLM | 21.16 | 81.7 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 29.3 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 26.7 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 31.37 | 12.2 | Meets every Standard requirement |
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 28.3 |
| Tag handling | 1.3 |
| Tone of voice | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Every LLM solution on the shortlist meets valid JSON and spacing at tag boundaries in Turkish. None of the three on LLM-based MT engines meet them. Besides the three that meet every Standard requirement, the solutions on Translated Lara 2 and OpenAI GPT-4.1 meet tag handling, but not general translation quality. With one-step solutions alone, Turkish is in tier 1 (simple solution + backup) for the Standard requirements, as with two-step solutions (translation + post-edit) and three-step solutions (translation + LQA + repair).
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Turkish | Class |
|---|---|---|
| Google Translation LLM | 81.7 | LLM-based MT engines |
| Anthropic Claude Sonnet 4.6 | 49.1 | LLMs |
| Google Gemini 3.5 Flash | 31.3 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 30.3 | LLMs |
| OpenAI GPT-4.1 | 29.7 | LLMs |
| Translated Lara 2 | 29.3 | LLM-based MT engines |
| Google Gemma 4 | 28.3 | Open-weight LLMs |
| DeepL Next Gen | 26.7 | LLM-based MT engines |
| Google Gemini 3.1 Pro Preview | 18.8 | LLMs |
| Anthropic Claude Opus 4.8 | 15.8 | LLMs |
| OpenAI GPT-5.6 Sol | 15.8 | LLMs |
| Anthropic Claude Fable 5 | 12.2 | LLMs |
| OpenAI GPT-5.5 | 12.2 | LLMs |
No translation into Turkish: Widn AI Tower Anthill (far too short, wrong language), Widn AI Tower Vesuvius (far too short, wrong language).
Every solution that meets general translation quality in Turkish is on an LLM. Translations from those five have 12.2 to 18.8 weighted errors per 1,000 words, from the other five LLMs 28.3 to 49.1, and from the three LLM-based MT engines 26.7 to 81.7. All ten LLMs on the shortlist support Turkish by our measured criteria.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Sonnet 5 | 0.793 | 10 | 1.96 | 16 | No | Not shortlisted |
| z.ai GLM-5 | 0.790 | 13 | 1.98 | 17 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.789 | 14 | 1.95 | 15 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.788 | 16 | 1.98 | 18 | No | Not shortlisted |
| DeepSeek V3.2 | 0.788 | 17 | 1.92 | 11 | No | Not shortlisted |
| Cohere Command A Translate | 0.787 | 19 | 1.94 | 13 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.786 | 20 | 2.07 | 20 | No | Not shortlisted |
| Mistral Large 3 | 0.783 | 21 | 1.92 | 12 | No | Not shortlisted |
| MiniMax M2.5 | 0.781 | 22 | 2.14 | 21 | No | Not shortlisted |
| Microsoft Translator | 0.775 | 23 | 2.33 | 25 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.771 | 24 | 2.26 | 23 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.771 | 25 | 2.37 | 27 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.770 | 26 | 2.53 | 29 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.769 | 27 | 2.39 | 28 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.769 | 28 | 2.21 | 22 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.768 | 29 | 2.60 | 31 | No | Not shortlisted |
| Cohere Tiny Aya | 0.762 | 30 | 2.66 | 34 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.760 | 31 | 2.36 | 26 | No | Not shortlisted |
| Supertext LLM | 0.758 | 32 | 2.55 | 30 | No | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | 0.757 | 33 | 2.61 | 32 | No | Not shortlisted |
| NiuTrans | 0.756 | 34 | 2.85 | 36 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.755 | 35 | 2.62 | 33 | No | Not shortlisted |
| Amazon Translate | 0.751 | 36 | 3.13 | 38 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.739 | 37 | 2.93 | 37 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.736 | 38 | 2.73 | 35 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.721 | 39 | 3.57 | 39 | No | Not shortlisted |
| Tencent Cloud TMT | 0.718 | 40 | 3.78 | 40 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.674 | 41 | 4.61 | 41 | No | Not shortlisted |
| AppTek NMT | 0.536 | 42 | 8.03 | 42 | No | Not shortlisted |
| DeepL Next Gen | 0.805 | 1 | 1.71 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.799 | 2 | 1.73 | 2 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | 0.799 | 3 | 1.75 | 4 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.797 | 4 | 1.90 | 10 | Yes | Top 8 by COMET-Kiwi |
| Google Gemma 4 | 0.796 | 5 | 1.73 | 3 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.796 | 6 | 1.94 | 14 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | 0.796 | 7 | 1.82 | 6 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.794 | 8 | 1.89 | 9 | Yes | Top 8 by COMET-Kiwi |
| xAI Grok 4.20 (non-reasoning) | 0.793 | 9 | 1.81 | 5 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.791 | 11 | 2.06 | 19 | Yes | Real-time place |
| OpenAI GPT-4.1 | 0.790 | 12 | 1.89 | 8 | Yes | Top 8 by MetricX |
| Anthropic Claude Sonnet 4.6 | 0.789 | 15 | 1.83 | 7 | Yes | Top 8 by MetricX |
| Google Translation LLM | 0.787 | 18 | 2.27 | 24 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Off-the-shelf NMT | 1 | 0.805 | 1 | 1.71 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 2 | 0.799 | 2 | 1.73 | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 3 | 0.799 | 4 | 1.75 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 4 | 0.797 | 10 | 1.90 | Top 8 by COMET-Kiwi |
| Google Gemma 4 | Adjusted LLM | 5 | 0.796 | 3 | 1.73 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 6 | 0.796 | 14 | 1.94 | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 7 | 0.796 | 6 | 1.82 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 8 | 0.794 | 9 | 1.89 | Top 8 by COMET-Kiwi |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 9 | 0.793 | 5 | 1.81 | Top 8 by MetricX |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 10 | 0.793 | 16 | 1.96 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 11 | 0.791 | 19 | 2.06 | Real-time place |
| OpenAI GPT-4.1 | Adjusted LLM | 12 | 0.790 | 8 | 1.89 | Top 8 by MetricX |
| z.ai GLM-5 | Adjusted LLM | 13 | 0.790 | 17 | 1.98 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.789 | 15 | 1.95 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 15 | 0.789 | 7 | 1.83 | Top 8 by MetricX |
| Mistral Medium 3.1 | Adjusted LLM | 16 | 0.788 | 18 | 1.98 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 17 | 0.788 | 11 | 1.92 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 18 | 0.787 | 24 | 2.27 | Real-time place |
| Cohere Command A Translate | Adjusted LLM | 19 | 0.787 | 13 | 1.94 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 20 | 0.786 | 20 | 2.07 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 21 | 0.783 | 12 | 1.92 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 22 | 0.781 | 21 | 2.14 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 23 | 0.775 | 25 | 2.33 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 24 | 0.771 | 23 | 2.26 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 25 | 0.771 | 27 | 2.37 | Not shortlisted |
| Tilde TildeOpen LLM engine | Off-the-shelf NMT | 26 | 0.770 | 29 | 2.53 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 27 | 0.769 | 28 | 2.39 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 28 | 0.769 | 22 | 2.21 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 29 | 0.768 | 31 | 2.60 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 30 | 0.762 | 34 | 2.66 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 31 | 0.760 | 26 | 2.36 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 32 | 0.758 | 30 | 2.55 | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | Adjusted LLM | 33 | 0.757 | 32 | 2.61 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 34 | 0.756 | 36 | 2.85 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 35 | 0.755 | 33 | 2.62 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 36 | 0.751 | 38 | 3.13 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 37 | 0.739 | 37 | 2.93 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 38 | 0.736 | 35 | 2.73 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 39 | 0.721 | 39 | 3.57 | Not shortlisted |
| Tencent Cloud TMT | Off-the-shelf NMT | 40 | 0.718 | 40 | 3.78 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 41 | 0.674 | 41 | 4.61 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 42 | 0.536 | 42 | 8.03 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.795 | — | 1.91 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.789 | — | 1.83 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.793 | — | 1.91 | Same model family as xAI Grok 4.20 (non-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 42 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 three solutions on the shortlist through COMET-Kiwi alone, on Anthropic Claude Opus 4.8, OpenAI GPT-5.5 and Anthropic Claude Fable 5, all meet general translation quality. None of the three through MetricX alone meet it. The two solutions with the fewest weighted errors, on Anthropic Claude Fable 5 and OpenAI GPT-5.5, rank 8 and 6 on COMET-Kiwi, and 9 and 14 on MetricX. The models of the tail nearest the cut are Anthropic Claude Sonnet 5, at rank 10 on COMET-Kiwi, and DeepSeek V3.2, at rank 11 on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $4.23 |
| Standard | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $4.23 |
| Advanced | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $10.44 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.13 | Open-weight LLMs |
| xAI Grok 4.20 (non-reasoning) | $1.03 | LLMs |
| OpenAI GPT-4.1 | $2.92 | LLMs |
| Google Gemini 3.5 Flash | $3.14 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.23 | LLMs |
| OpenAI GPT-5.6 Sol | $7.23 | LLMs |
| Anthropic Claude Sonnet 4.6 | $7.54 | LLMs |
| OpenAI GPT-5.5 | $10.44 | LLMs |
| Anthropic Claude Opus 4.8 | $15.65 | LLMs |
| Google Translation LLM | $21.16 | 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 | $31.37 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Fable 5 | 0.52 | LLMs |
| Anthropic Claude Opus 4.8 | 0.519 | LLMs |
| Anthropic Claude Sonnet 5 | 0.519 | LLMs |
| Mistral Medium 3.1 | 0.478 | LLMs |
| Mistral Large 3 | 0.475 | LLMs |
| AI21 Jamba Large 1.7 | 0.441 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.421 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.418 | LLMs |
| DeepSeek V3.2 | 0.396 | LLMs |
| MoonshotAI Kimi K2.5 | 0.381 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.329 | Open-weight LLMs |
| z.ai GLM-5 | 0.327 | LLMs |
| MiniMax M2.5 | 0.313 | LLMs |
| OpenAI GPT-4.1 | 0.293 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.291 | LLMs |
| OpenAI GPT-5.6 Luna | 0.291 | LLMs |
| OpenAI GPT-5.6 Sol | 0.291 | LLMs |
| OpenAI GPT-5.6 Terra | 0.291 | LLMs |
| Google Gemini 3.5 Flash | 0.29 | LLMs |
| OpenAI GPT-5.5 | 0.29 | LLMs |
| Google Gemma 4 | 0.289 | Open-weight LLMs |
| Cohere Command A Translate | 0.284 | LLMs |
| Cohere Tiny Aya | 0.276 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.274 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.274 | LLMs |
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 45 |
| Standard | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 45 |
| Advanced | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 45 |
| 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 |
| Pangeanic PangeaMT | 0.6 | MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| Amazon Translate | 1.2 | MT engines |
| SYSTRAN PNMT API | 1.2 | MT engines |
| Translated Lara 2 | 1.4 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 1.9 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.7 | MT engines |
| Oracle Cloud Language Service | 4.1 | MT engines |
| Tilde TildeOpen LLM engine | 8.4 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 15 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 16 | LLMs |
| Supertext LLM | 16 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 16 | LLMs |
| Cohere Tiny Aya | 19 | LLMs |
| Alibaba Machine Translation API | 19 | MT engines |
| OpenAI GPT-4.1 | 20 | LLMs |
| OpenAI GPT-5.6 Terra | 21 | LLMs |
| Anthropic Claude Haiku 4.5 | 38 | LLMs |
| Cohere Command A Translate | 39 | LLMs |
| DeepSeek V3.2 | 40 | LLMs |
| MiniMax M2.5 | 42 | LLMs |
| Swiss AI Apertus-8B-2509 | 44 | Open-weight LLMs |
| OpenAI GPT-5.5 | 45 | LLMs |
| Mistral Medium 3.1 | 47 | LLMs |
| xAI Grok 4.20 (reasoning) | 48 | LLMs |
| OpenAI GPT-5.6 Sol | 50 | LLMs |
| Anthropic Claude Sonnet 5 | 51 | LLMs |
| NiuTrans | 55 | MT engines |
| Mistral Large 3 | 56 | LLMs |
| Anthropic Claude Fable 5 | 57 | LLMs |
| Tencent Cloud TMT | 58 | MT engines |
| AI21 Jamba Large 1.7 | 60 | LLMs |
| Anthropic Claude Opus 4.8 | 63 | LLMs |
| SpeakLeash Bielik-11B-v3.0-Instruct | 69 | Open-weight LLMs |
| Anthropic Claude Sonnet 4.6 | 74 | LLMs |
| Google Gemini 3.1 Pro Preview | 98 | LLMs |
| AppTek NMT | 110 | MT engines |
| Alibaba Qwen3-235B-A22B | 136 | Open-weight LLMs |
| Google Gemma 4 | 199 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 213 | LLMs |
| z.ai GLM-5 | 241 | 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 Turkish.
| 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 | one-step | 100% | ✓ | $31.37 | 57 |
| OpenAI GPT-5.5 | one-step | 100% | ✓ | $10.44 | 45 |
For tone of voice in Turkish we check the formal register, with siz as the form of address. For gender neutrality we check that a translation does not add a gender the source does not specify, and a default gender is never allowed. Turkish has no grammatical gender for persons, so a translation is neutral by default and the requirement comes down to avoiding gendered terms. Translations from both of the two best solutions are gender-neutral in 100% of the sentences about people, so the gender-neutrality agent is not needed.
For each pair of automatic metrics, the whole confidence interval of the agreement in Turkish is above the threshold, so neither pair is borderline. Across the 13 shortlisted solutions, the median share of sentences where the three LQA judges agree is 94.9%. So both the shortlist and the Intento LQA results for Turkish carry full weight.
Leave feedback