The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Burmese 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 Burmese. |
| Tone of voice | The formal register (honorific vocabulary) 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 Burmese, 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.
The best available solution for the Basic and Standard requirements and the cheapest solution with a conclusive Intento LQA result differ in total errors, cost and time. The three-step solution (translation + LQA + repair) on OpenAI GPT-5.5 has 14.9 total errors per 1,000 words, at $681 per million characters and 674 seconds per 1,000 words. The one-step solution on OpenAI GPT-5.6 Luna has 35.3 total errors per 1,000 words, at $0.77 per million characters and 27 seconds per 1,000 words. Both miss general translation quality.
A limit on the steps leaves more errors. With at most two steps, the fewest total errors are 29.4 per 1,000 words, from the two-step solution (translation + post-edit) on OpenAI GPT-5.5, at $54.57 per million characters and 530 seconds per 1,000 words. The two-step solution on OpenAI GPT-5.6 Luna has 35.0 total errors per 1,000 words, at $1.76 per million characters and 53 seconds per 1,000 words.
The gender-neutrality agent works with the three-step solution on OpenAI GPT-5.5 for the Advanced requirements. The solution then costs $1,132 per million characters instead of $681, and takes 730 seconds per 1,000 words instead of 674. It has 15.8 total errors per 1,000 words, against 14.9 without the agent, and meets gender neutrality. With one critical error, it still misses general translation quality.
OpenAI GPT-5.5
three-step
Misses general translation quality.
OpenAI GPT-5.5
three-step
Misses general translation quality.
OpenAI GPT-5.5
three-step with the gender-neutrality agent
Misses general translation quality.
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 4: no solution | No solution | — | — | — | ||
| Standard | Tier 4: no solution | No solution | — | — | — | ||
| Advanced | Tier 4: no solution | No solution | — | — | — |
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.
Each level up adds requirements that the best available solution meets. Under the Basic and Standard requirements it is the same three-step solution (translation + LQA + repair) on OpenAI GPT-5.5, and its translations meet tone of voice, tag handling and formatting. For the Advanced requirements the gender-neutrality agent works with it, and it then meets gender neutrality. Of the requirements of all three levels, it misses only general translation quality. As no solution meets any level, Optimize for has nothing to select.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-5.6 Luna | 0.77 | 33.3 | Misses a Standard requirement |
| DeepSeek V3.2 | 2.17 | 59 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 5.04 | 54.1 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 6.52 | 41.5 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 19.83 | 33.6 | Misses a Standard requirement |
| Google Translation LLM | 24.64 | 112.7 | Misses a Standard requirement |
Among the one-step solutions with a conclusive result, none that costs more than the one on OpenAI GPT-5.6 Luna has fewer weighted errors. The solution on OpenAI GPT-5.5 costs $19.83 per million characters for 33.6 weighted errors per 1,000 words, and the one on Google Translation LLM $24.64, the most, for 112.7. The solutions on DeepSeek V3.2, Google Gemini 3.5 Flash and Google Gemini 3.1 Pro Preview cost $2.17 to $6.52 per million characters and have 41.5 to 59.0 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 47.8 |
| Tag handling | 1 |
| Gender neutrality | 0.3 |
| Tone of voice | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
The one-step solutions with a conclusive result have 33.3 to 112.7 weighted errors per 1,000 words and 1 to 16 critical errors each. With three steps (translation + LQA + repair), the weighted errors come down to 13.2 to 28.0 per 1,000 words, within the bar only on OpenAI GPT-5.5, and each solution still has more critical or major errors than the requirement allows. With two or three steps, translations from DeepSeek V3.2 have 100% of the tags right, against 97.6% with one step.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Burmese | Class |
|---|---|---|
| Google Translation LLM | 112.7 | LLM-based MT engines |
| DeepSeek V3.2 | 59 | LLMs |
| Google Gemini 3.5 Flash | 54.1 | LLMs |
| Google Gemini 3.1 Pro Preview | 41.5 | LLMs |
| OpenAI GPT-5.5 | 33.6 | LLMs |
| OpenAI GPT-5.6 Luna | 33.3 | LLMs |
No translation into Burmese: AI21 Jamba Large 1.7 (cut off part-way, garbage or near-empty, wrong language, wrong script), Alibaba Machine Translation API (cut off part-way).
Translations from the five LLMs with a conclusive Intento LQA result have 33.3 to 59.0 weighted errors per 1,000 words, and those from Google Translation LLM, the one LLM-based MT engine with a conclusive result, 112.7. None of the five MT engines that are not LLM-based made the shortlist, and the highest, Microsoft Translator, is at rank 17 on COMET-Kiwi and 15 on MetricX. The open-weight Google Gemma 4 and the LLM-based MT engines DeepL Next Gen and Translated Lara 2 are on the shortlist with no conclusive result.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| OpenAI GPT-4.1 | 0.772 | 10 | 1.87 | 9 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.768 | 12 | 2.16 | 13 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.756 | 14 | 2.14 | 12 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.753 | 15 | 2.53 | 17 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.750 | 16 | 2.48 | 16 | No | Not shortlisted |
| Microsoft Translator | 0.749 | 17 | 2.41 | 15 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.740 | 18 | 2.88 | 19 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.735 | 19 | 2.98 | 21 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.734 | 20 | 3.01 | 22 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.729 | 21 | 2.94 | 20 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.728 | 22 | 3.16 | 24 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.726 | 23 | 2.59 | 18 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.725 | 24 | 3.22 | 25 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.724 | 25 | 3.06 | 23 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.701 | 26 | 3.47 | 26 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.678 | 27 | 3.70 | 27 | No | Not shortlisted |
| Cohere Tiny Aya | 0.652 | 28 | 4.55 | 28 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.586 | 29 | 7.33 | 29 | No | Not shortlisted |
| MiniMax M2.5 | 0.453 | 30 | 10.91 | 30 | No | Not shortlisted |
| Cohere Command A Translate | 0.341 | 31 | 14.79 | 31 | No | Not shortlisted |
| OpenAI GPT-5.6 Luna | 0.788 | 1 | 1.50 | 1 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.786 | 2 | 1.78 | 6 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.785 | 3 | 1.70 | 5 | Yes | Top 8 by both metrics |
| Google Gemma 4 | 0.782 | 4 | 1.68 | 4 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.777 | 5 | 1.64 | 3 | Yes | Top 8 by both metrics |
| DeepSeek V3.2 | 0.775 | 6 | 1.54 | 2 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.775 | 7 | 1.99 | 10 | Yes | Top 8 by COMET-Kiwi |
| Mistral Large 3 | 0.774 | 8 | 1.81 | 7 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.774 | 9 | 1.82 | 8 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.772 | 11 | 2.08 | 11 | Yes | Real-time place |
| Google Translation LLM | 0.767 | 13 | 2.29 | 14 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Luna | Adjusted LLM | 1 | 0.788 | 1 | 1.50 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 2 | 0.786 | 6 | 1.78 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 3 | 0.785 | 5 | 1.70 | Top 8 by both metrics |
| Google Gemma 4 | Adjusted LLM | 4 | 0.782 | 4 | 1.68 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 5 | 0.777 | 3 | 1.64 | Top 8 by both metrics |
| DeepSeek V3.2 | Adjusted LLM | 6 | 0.775 | 2 | 1.54 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 7 | 0.775 | 10 | 1.99 | Top 8 by COMET-Kiwi |
| Mistral Large 3 | Adjusted LLM | 8 | 0.774 | 7 | 1.81 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 9 | 0.774 | 8 | 1.82 | Top 8 by MetricX |
| OpenAI GPT-4.1 | Adjusted LLM | 10 | 0.772 | 9 | 1.87 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 11 | 0.772 | 11 | 2.08 | Real-time place |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 12 | 0.768 | 13 | 2.16 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 13 | 0.767 | 14 | 2.29 | Real-time place |
| OpenAI gpt-oss-120b | Adjusted LLM | 14 | 0.756 | 12 | 2.14 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 15 | 0.753 | 17 | 2.53 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 16 | 0.750 | 16 | 2.48 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 17 | 0.749 | 15 | 2.41 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 18 | 0.740 | 19 | 2.88 | Not shortlisted |
| Anthropic Claude Opus 4.8 | Off-the-shelf LLM | 19 | 0.735 | 21 | 2.98 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 20 | 0.734 | 22 | 3.01 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 21 | 0.729 | 20 | 2.94 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 22 | 0.728 | 24 | 3.16 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 23 | 0.726 | 18 | 2.59 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 24 | 0.725 | 25 | 3.22 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 25 | 0.724 | 23 | 3.06 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 26 | 0.701 | 26 | 3.47 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 27 | 0.678 | 27 | 3.70 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 28 | 0.652 | 28 | 4.55 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 29 | 0.586 | 29 | 7.33 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 30 | 0.453 | 30 | 10.91 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 31 | 0.341 | 31 | 14.79 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.786 | — | 1.50 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.782 | — | 1.82 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.763 | — | 2.11 | 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 31 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.
Intento LQA and the automatic metrics rank the shortlist in a different order. OpenAI GPT-5.5 made it through MetricX alone, at rank 8, and DeepL Next Gen through COMET-Kiwi alone, at rank 7. Translations from OpenAI GPT-5.5 have 33.6 weighted errors per 1,000 words, against 33.3 for OpenAI GPT-5.6 Luna, and translations from DeepSeek V3.2, second on MetricX, have 59.0. Nearest the cut in the tail is OpenAI GPT-4.1, at rank 10 on COMET-Kiwi and rank 9 on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | No solution | — |
| Standard | No solution | — |
| Advanced | No solution | — |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.19 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | $0.77 | LLMs |
| DeepSeek V3.2 | $2.17 | LLMs |
| Mistral Large 3 | $2.51 | LLMs |
| Google Gemini 3.5 Flash | $5.04 | LLMs |
| z.ai GLM-5 | $5.23 | LLMs |
| Google Gemini 3.1 Pro Preview | $6.52 | LLMs |
| OpenAI GPT-5.5 | $19.83 | LLMs |
| Google Translation LLM | $24.64 | LLM-based MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| MiniMax M2.5 | 1.675 | LLMs |
| Cohere Command A Translate | 1.327 | LLMs |
| Alibaba Qwen3-235B-A22B | 1.269 | Open-weight LLMs |
| Mistral Large 3 | 1.25 | LLMs |
| Mistral Medium 3.1 | 1.21 | LLMs |
| Anthropic Claude Opus 4.8 | 0.912 | LLMs |
| Anthropic Claude Fable 5 | 0.911 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.903 | LLMs |
| Anthropic Claude Sonnet 5 | 0.9 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.885 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.879 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.873 | LLMs |
| z.ai GLM-5 | 0.835 | LLMs |
| DeepSeek V3.2 | 0.733 | LLMs |
| MoonshotAI Kimi K2.5 | 0.513 | LLMs |
| OpenAI GPT-5.6 Sol | 0.47 | LLMs |
| OpenAI GPT-5.6 Luna | 0.468 | LLMs |
| OpenAI GPT-5.6 Terra | 0.468 | LLMs |
| OpenAI GPT-5.5 | 0.461 | LLMs |
| OpenAI GPT-4.1 | 0.46 | LLMs |
| Cohere Tiny Aya | 0.391 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.37 | LLMs |
| Google Gemini 3.5 Flash | 0.367 | LLMs |
| Google Gemma 4 | 0.366 | Open-weight LLMs |
The tokenizers of 24 models produce a median 0.85 tokens per character of the translation, 2.6 times as many as for English. Tokens per character do not set the order of the costs alone. With 1.25 tokens per character, the solution on Mistral Large 3 costs $2.51 per million characters, and with 0.367, the one on Google Gemini 3.5 Flash costs $5.04. On OpenAI GPT-5.5, one step costs $19.83 per million characters, two steps $54.57, three steps $681 and three steps with the gender-neutrality agent $1,132.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | No solution | — |
| Standard | No solution | — |
| Advanced | No solution | — |
| 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 | 0.4 | MT engines |
| Google Translation LLM | 0.8 | LLM-based MT engines |
| DeepL Next Gen | 0.9 | LLM-based MT engines |
| SYSTRAN PNMT API | 1.7 | MT engines |
| Translated Lara 2 | 3.5 | LLM-based MT engines |
| Yandex Cloud Translate v2 | 3.6 | MT engines |
| OpenAI gpt-oss-120b | 4.1 | Open-weight LLMs |
| Cohere Tiny Aya | 24 | LLMs |
| Google Gemini 3.5 Flash | 25 | LLMs |
| OpenAI GPT-5.6 Luna | 27 | LLMs |
| OpenAI GPT-4.1 | 45 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 53 | LLMs |
| DeepSeek V3.2 | 65 | LLMs |
| OpenAI GPT-5.6 Terra | 73 | LLMs |
| Anthropic Claude Haiku 4.5 | 75 | LLMs |
| Swiss AI Apertus-8B-2509 | 84 | Open-weight LLMs |
| Mistral Medium 3.1 | 92 | LLMs |
| xAI Grok 4.20 (reasoning) | 108 | LLMs |
| Google Gemini 3.1 Pro Preview | 111 | LLMs |
| Cohere Command A Translate | 112 | LLMs |
| Anthropic Claude Sonnet 5 | 116 | LLMs |
| Mistral Large 3 | 129 | LLMs |
| Google Gemma 4 | 174 | Open-weight LLMs |
| Anthropic Claude Sonnet 4.6 | 178 | LLMs |
| Anthropic Claude Fable 5 | 204 | LLMs |
| Anthropic Claude Opus 4.8 | 245 | LLMs |
| MiniMax M2.5 | 273 | LLMs |
| OpenAI GPT-5.5 | 284 | LLMs |
| MoonshotAI Kimi K2.5 | 333 | LLMs |
| OpenAI GPT-5.6 Sol | 364 | LLMs |
| z.ai GLM-5 | 398 | LLMs |
| Alibaba Qwen3-235B-A22B | 472 | 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 Burmese.
Of the eight real-time models in Burmese, four are MT engines, three LLM-based MT engines and one the open-weight LLM OpenAI gpt-oss-120b, at 4.1 seconds per 1,000 words the fastest LLM. Only the LLM-based MT engines, Google Translation LLM, DeepL Next Gen and Translated Lara 2, are on the shortlist. The solution on Google Translation LLM misses tag handling and formatting, and the other two have no conclusive Intento LQA result. The best available solution takes 674 seconds per 1,000 words, and 730 with the gender-neutrality agent.
| 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 | three-step, with the gender-neutrality agent | 100% | ✗ | $1,132 | 730 |
| DeepSeek V3.2 | three-step, with the gender-neutrality agent | 100% | ✗ | $1,109 | 341 |
For tone of voice we check the formal register, with honorific vocabulary. Burmese has no grammatical gender for persons, so translations are gender-neutral unless they use gendered terms. With the gender-neutrality agent, both best solutions meet gender neutrality. The three-step solution (translation + LQA + repair) on DeepSeek V3.2 then costs $1,109 per million characters instead of $649, and has 27.7 total errors per 1,000 words instead of 22.1, with no critical error but five major errors.
The metric agreement in Burmese is above the threshold and not borderline, so the automatic ranks behind the shortlist carry weight. Across the 11 shortlisted solutions, the median agreement of the three LQA judges is 92.1% of the sentences, and there is no conclusive result for Google Gemma 4, Translated Lara 2, z.ai GLM-5, DeepL Next Gen and Mistral Large 3. The general translation quality results on this page rest on the other six one-step solutions, and on the two-step and three-step solutions on the five LLMs among them.
Leave feedback