The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Sinhala 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 Sinhala. |
| Tone of voice | The formal register (ඔබ) 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 Sinhala, 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. 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.
Two solutions meet every Standard requirement in Sinhala, and neither leads on errors, cost and time at once. The three-step solution (translation + LQA + repair) on OpenAI GPT-5.6 Luna has 14.3 total errors per 1,000 words against 20.9 for the two-step solution (translation + post-edit) on Anthropic Claude Fable 5. It takes 136 seconds per 1,000 words against 247, and costs $556 per million characters against $46.00. Optimize for selects OpenAI GPT-5.6 Luna for quality and latency, and Anthropic Claude Fable 5 for cost.
With one step only, the solutions on Google Gemini 3.1 Pro Preview, at $5.77 per million characters and 82 seconds per 1,000 words, and Anthropic Claude Fable 5, at $46.00 and 247 seconds, meet every Basic requirement. None of the real-time solutions meet even the Basic requirements.
For the Advanced requirements, the gender-neutrality agent works with both solutions that meet every Standard requirement. On OpenAI GPT-5.6 Luna it raises the cost from $556 to $926 per million characters and the time from 136 to 147 seconds per 1,000 words, and total errors fall from 14.3 to 10.6 per 1,000 words. On Anthropic Claude Fable 5 the cost goes from $46.00 to $93.29 per million characters, the time from 247 to 514 seconds per 1,000 words and total errors from 20.9 to 17.5 per 1,000 words.
OpenAI GPT-5.6 Luna
three-step
Google Gemini 3.1 Pro Preview
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
three-step
Anthropic Claude Fable 5
two-step
OpenAI GPT-5.6 Luna
three-step
OpenAI GPT-5.6 Luna
three-step with the gender-neutrality agent
Misses tone of voice, gender neutrality.
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 | Google Gemini 3.1 Pro Preview | 23.6 | $5.77 | 82 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | two-step | Anthropic Claude Fable 5 | 20.9 | $46 | 247 |
| 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.
Under the Basic requirements, the solutions on Google Gemini 3.1 Pro Preview and Anthropic Claude Fable 5 take one step, and the other three are three-step solutions (translation + LQA + repair). The Standard requirements remove three of the five models on tone of voice and take Anthropic Claude Fable 5 to two steps (translation + post-edit). Optimize for selects OpenAI GPT-5.6 Luna for quality under both sets of requirements, and Google Gemini 3.1 Pro Preview for cost and latency under the Basic ones.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.17 | 60 | Misses a Standard requirement |
| OpenAI GPT-5.6 Luna | 0.63 | 29.7 | Misses a Standard requirement |
| DeepSeek V3.2 | 1.87 | 53.4 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 4.37 | 34.3 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.77 | 19.8 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 15.62 | 31.6 | Misses a Standard requirement |
| Google Cloud Advanced Translation API | 20 | 119 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 22.88 | 32.3 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 87.7 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 46 | 17.1 | Misses a Standard requirement |
Of the shortlisted one-step solutions in Sinhala with a conclusive Intento LQA result, a higher price comes with fewer weighted errors only along four of them: from the open-weight Google Gemma 4, the cheapest, at $0.17 per million characters and 60.0 weighted errors per 1,000 words, through the LLMs OpenAI GPT-5.6 Luna at $0.63 and 29.7 and Google Gemini 3.1 Pro Preview at $5.77 and 19.8, to Anthropic Claude Fable 5 at $46.00 and 17.1. Each of the others has more weighted errors than a cheaper one.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 33.3 |
| Gender neutrality | 3.3 |
| Tone of voice | 0.1 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
General translation quality holds back most solutions in Sinhala. Besides the two that meet it, the one-step solutions with a conclusive result have 29.7 to 119.0 weighted errors per 1,000 words. Three-step solutions (translation + LQA + repair) bring OpenAI GPT-5.6 Luna, DeepSeek V3.2 and Google Gemini 3.5 Flash under the bar, at 11.2, 13.5 and 17.8 weighted errors per 1,000 words. The one-step solution on Translated Lara 2 meets tone of voice and tag handling, and the one on Google Cloud Advanced Translation API meets neither.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Sinhala | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 119 | MT engines |
| Translated Lara 2 | 87.7 | LLM-based MT engines |
| Google Gemma 4 | 60 | Open-weight LLMs |
| DeepSeek V3.2 | 53.4 | LLMs |
| Google Gemini 3.5 Flash | 34.3 | LLMs |
| Anthropic Claude Opus 4.8 | 32.3 | LLMs |
| OpenAI GPT-5.5 | 31.6 | LLMs |
| OpenAI GPT-5.6 Luna | 29.7 | LLMs |
| Google Gemini 3.1 Pro Preview | 19.8 | LLMs |
| Anthropic Claude Fable 5 | 17.1 | LLMs |
No translation into Sinhala: AI21 Jamba Large 1.7 (cut off part-way, garbage or near-empty, wrong language, wrong script).
In one step, translations from the eight shortlisted LLMs with a conclusive Intento LQA result have 17.1 to 60.0 weighted errors per 1,000 words, the most from the open-weight Google Gemma 4. Translations from Translated Lara 2, an LLM-based MT engine, have 87.7 weighted errors per 1,000 words and 29 major errors, and those from the MT engine Google Cloud Advanced Translation API 119.0, with 8 critical errors, the only ones in the 10 one-step solutions with a conclusive result. All eight LLMs support Sinhala by our measured criteria.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Sonnet 5 | 0.850 | 10 | 1.55 | 10 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.847 | 12 | 1.77 | 13 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.846 | 13 | 1.57 | 11 | No | Not shortlisted |
| xAI Grok 4.20 (non-reasoning) | 0.841 | 14 | 1.77 | 14 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.836 | 15 | 1.82 | 15 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.801 | 18 | 2.75 | 20 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.798 | 19 | 2.45 | 17 | No | Not shortlisted |
| Amazon Translate | 0.794 | 20 | 2.52 | 18 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.787 | 21 | 2.83 | 21 | No | Not shortlisted |
| MiniMax M2.5 | 0.769 | 22 | 3.56 | 23 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.753 | 23 | 3.53 | 22 | No | Not shortlisted |
| Mistral Large 3 | 0.724 | 24 | 3.71 | 24 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.702 | 25 | 4.16 | 25 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.467 | 26 | 11.51 | 27 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.454 | 27 | 10.55 | 26 | No | Not shortlisted |
| Cohere Command A Translate | 0.411 | 28 | 13.07 | 28 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.382 | 29 | 13.36 | 29 | No | Not shortlisted |
| Google Gemini 3.5 Flash | 0.874 | 1 | 1.13 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.874 | 2 | 1.14 | 2 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.871 | 3 | 1.27 | 4 | Yes | Top 8 by both metrics |
| Google Gemma 4 | 0.869 | 4 | 1.27 | 3 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.865 | 5 | 1.50 | 9 | Yes | Top 8 by COMET-Kiwi |
| z.ai GLM-5 | 0.862 | 6 | 1.38 | 6 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.860 | 7 | 1.29 | 5 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.858 | 8 | 1.39 | 7 | Yes | Top 8 by both metrics |
| Translated Lara 2 | 0.851 | 9 | 1.58 | 12 | Yes | Real-time place |
| DeepSeek V3.2 | 0.848 | 11 | 1.44 | 8 | Yes | Top 8 by MetricX |
| Microsoft Translator | 0.817 | 16 | 1.92 | 16 | Yes | Real-time place |
| Google Cloud Advanced Translation API | 0.814 | 17 | 2.56 | 19 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Google Gemini 3.5 Flash | Adjusted LLM | 1 | 0.874 | 1 | 1.13 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 2 | 0.874 | 2 | 1.14 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 3 | 0.871 | 4 | 1.27 | Top 8 by both metrics |
| Google Gemma 4 | Adjusted LLM | 4 | 0.869 | 3 | 1.27 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 5 | 0.865 | 9 | 1.50 | Top 8 by COMET-Kiwi |
| z.ai GLM-5 | Adjusted LLM | 6 | 0.862 | 6 | 1.38 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 7 | 0.860 | 5 | 1.29 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 8 | 0.858 | 7 | 1.39 | Top 8 by both metrics |
| Translated Lara 2 | Adjusted NMT | 9 | 0.851 | 12 | 1.58 | Real-time place |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 10 | 0.850 | 10 | 1.55 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 11 | 0.848 | 8 | 1.44 | Top 8 by MetricX |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 12 | 0.847 | 13 | 1.77 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 13 | 0.846 | 11 | 1.57 | Not shortlisted |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 14 | 0.841 | 14 | 1.77 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 15 | 0.836 | 15 | 1.82 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 16 | 0.817 | 16 | 1.92 | Real-time place |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 17 | 0.814 | 19 | 2.56 | Real-time place |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 18 | 0.801 | 20 | 2.75 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 19 | 0.798 | 17 | 2.45 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 20 | 0.794 | 18 | 2.52 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 21 | 0.787 | 21 | 2.83 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 22 | 0.769 | 23 | 3.56 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Off-the-shelf LLM | 23 | 0.753 | 22 | 3.53 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 24 | 0.724 | 24 | 3.71 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 25 | 0.702 | 25 | 4.16 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 26 | 0.467 | 27 | 11.51 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 27 | 0.454 | 26 | 10.55 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 28 | 0.411 | 28 | 13.07 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 29 | 0.382 | 29 | 13.36 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.870 | — | 1.20 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.862 | — | 1.36 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.840 | — | 1.95 | 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 29 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.
COMET-Kiwi put OpenAI GPT-5.5 on the shortlist, fifth on COMET-Kiwi and ninth on MetricX, and MetricX put DeepSeek V3.2 there, eighth on MetricX and 11th on COMET-Kiwi. By Intento LQA, the one-step solutions on them have 31.6 and 53.4 weighted errors per 1,000 words, and the three-step solution (translation + LQA + repair) on DeepSeek V3.2 meets every Basic requirement. Nearest the cut in the tail is Anthropic Claude Sonnet 5, at rank 10 on both metrics.
| 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 | $5.77 |
| Standard | Anthropic Claude Fable 5, two-step | $46.00 |
| Advanced | No solution | — |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.17 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | $0.63 | LLMs |
| DeepSeek V3.2 | $1.87 | LLMs |
| Google Gemini 3.5 Flash | $4.37 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.77 | LLMs |
| z.ai GLM-5 | $6.39 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $15.62 | LLMs |
| Google Cloud Advanced Translation API | $20 | MT engines |
| Anthropic Claude Opus 4.8 | $22.88 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $46 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Cohere Command A Translate | 1.423 | LLMs |
| Mistral Large 3 | 1.366 | LLMs |
| z.ai GLM-5 | 1.364 | LLMs |
| Mistral Medium 3.1 | 1.349 | LLMs |
| Alibaba Qwen3-235B-A22B | 1.131 | Open-weight LLMs |
| MiniMax M2.5 | 1.03 | LLMs |
| Anthropic Claude Fable 5 | 0.837 | LLMs |
| Anthropic Claude Opus 4.8 | 0.835 | LLMs |
| Anthropic Claude Sonnet 5 | 0.822 | LLMs |
| OpenAI GPT-4.1 | 0.817 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.809 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.805 | LLMs |
| DeepSeek V3.2 | 0.771 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.738 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.738 | LLMs |
| MoonshotAI Kimi K2.5 | 0.7 | LLMs |
| OpenAI GPT-5.6 Sol | 0.489 | LLMs |
| OpenAI GPT-5.5 | 0.485 | LLMs |
| OpenAI GPT-5.6 Luna | 0.484 | LLMs |
| OpenAI GPT-5.6 Terra | 0.484 | LLMs |
| Google Gemma 4 | 0.416 | Open-weight LLMs |
| Google Gemini 3.1 Pro Preview | 0.415 | LLMs |
| Google Gemini 3.5 Flash | 0.413 | LLMs |
The tokenizers produce a median 0.81 tokens per character of Sinhala. On the same tokenizer, Sinhala has a median 3.1 times the tokens per character of the English source. The tokenizers of the two shortlisted Anthropic Claude models produce 0.835 and 0.837 tokens per character, those of the three shortlisted Google LLMs 0.413 to 0.416. The one-step solution on Google Gemini 3.1 Pro Preview costs $5.77 per million characters, the two-step one (translation + post-edit) $26.52 and the three-step one (translation + LQA + repair) $582.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | 82 |
| Standard | OpenAI GPT-5.6 Luna, three-step | 136 |
| 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 |
| Amazon Translate | 1.3 | MT engines |
| Yandex Cloud Translate v2 | 3.2 | MT engines |
| OpenAI gpt-oss-120b | 3.3 | Open-weight LLMs |
| Translated Lara 2 | 6.4 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 23 | LLMs |
| Alibaba Machine Translation API | 29 | MT engines |
| OpenAI GPT-4.1 | 44 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 47 | LLMs |
| Google Gemini 3.5 Flash | 67 | LLMs |
| OpenAI GPT-5.6 Terra | 72 | LLMs |
| Anthropic Claude Haiku 4.5 | 79 | LLMs |
| Google Gemini 3.1 Pro Preview | 82 | LLMs |
| DeepSeek V3.2 | 87 | LLMs |
| xAI Grok 4.20 (reasoning) | 100 | LLMs |
| Cohere Command A Translate | 126 | LLMs |
| MiniMax M2.5 | 127 | LLMs |
| Swiss AI Apertus-8B-2509 | 137 | Open-weight LLMs |
| Anthropic Claude Sonnet 4.6 | 148 | LLMs |
| Google Gemma 4 | 188 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 221 | LLMs |
| OpenAI GPT-5.5 | 222 | LLMs |
| OpenAI GPT-5.6 Sol | 226 | LLMs |
| Anthropic Claude Fable 5 | 247 | LLMs |
| Anthropic Claude Opus 4.8 | 275 | LLMs |
| Mistral Large 3 | 330 | LLMs |
| MoonshotAI Kimi K2.5 | 349 | LLMs |
| Alibaba Qwen3-235B-A22B | 403 | Open-weight LLMs |
| z.ai GLM-5 | 705 | LLMs |
| Mistral Medium 3.1 | 2,192 | 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 Sinhala.
| 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-5.6 Luna | three-step, with the gender-neutrality agent | 98.2% | ✗ | $926 | 147 |
| Anthropic Claude Fable 5 | two-step, with the gender-neutrality agent | 100% | ✗ | $93.29 | 514 |
For Sinhala we check the formal register, with ඔබ as the form of address, and gender-neutral wording through neutral nouns and rephrasing, with no default gender. With the gender-neutrality agent, the two best solutions have 97.9% of the sentences in the formal register, against 100% without, so neither meets tone of voice. The one on Anthropic Claude Fable 5, a two-step solution (translation + post-edit), does not meet tag handling either, with 100% of the sentences about people gender-neutral but 99.6% of the tags right.
Both pairs of automatic metrics agree in Sinhala above the threshold, and neither pair is borderline. The three LQA judges agree on a median of 94.7% of the sentences across the 12 shortlisted solutions. With agreement that high, the Intento LQA error counts on this page carry weight.
Leave feedback