The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Yiddish 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 Yiddish. |
| Tone of voice | The formal register (איר) 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 Yiddish, 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 three-step solutions (translation + LQA + repair) meet every Standard requirement in Yiddish, and neither leads on errors, cost and time at once. The one on Anthropic Claude Fable 5 has 10.9 total errors per 1,000 words and takes 408 seconds per 1,000 words, against 18.9 and 438 for the one on OpenAI GPT-5.5. The one on OpenAI GPT-5.5 costs less, $653 against $683 per million characters. Optimizing for quality or latency selects the solution on Anthropic Claude Fable 5, and optimizing for cost the one on OpenAI GPT-5.5.
None of the solutions that meet even the Basic requirements in Yiddish cost less than $652 per million characters or take less than 407 seconds per 1,000 words. Both are three-step solutions (translation + LQA + repair). The three-step solution on OpenAI GPT-5.6 Terra costs less and takes less time than both, at $640 per million characters and 307 seconds per 1,000 words. Its translations have 22.1 weighted errors per 1,000 words, over the bar.
For the Advanced requirements, the gender-neutrality agent works with both three-step solutions (translation + LQA + repair). On Anthropic Claude Fable 5 it raises the cost from $683 to $1,116 per million characters, and on OpenAI GPT-5.5 from $653 to $1,122. Time rises from 408 to 440 and from 438 to 473 seconds per 1,000 words. Total errors rise, from 10.9 to 11.8 and from 18.9 to 20.1 per 1,000 words. Weighted errors rise, from 8.9 to 11.5 and from 16.5 to 20.1 per 1,000 words, the last over the bar.
Anthropic Claude Fable 5
three-step
OpenAI GPT-5.5
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step
OpenAI GPT-5.5
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Misses 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 2: two or more solutions, none simple | ☆☆ | three-step | OpenAI GPT-5.5 | 18.9 | $653 | 438 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | three-step | OpenAI GPT-5.5 | 18.9 | $653 | 438 |
| 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.
The two three-step solutions (translation + LQA + repair) that meet every Basic requirement in Yiddish also meet every Standard requirement, and Optimize for selects the same solutions for both. None of the one-step or two-step solutions (translation + post-edit) meet even the Basic requirements. The Advanced requirements remove both three-step solutions, also with the gender-neutrality agent. The best available, on Anthropic Claude Fable 5 with the agent, has 11.8 total errors per 1,000 words and misses gender neutrality.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| xAI Grok 4.20 (non-reasoning) | 1.78 | 307.5 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.69 | 168.8 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 5.5 | 71.9 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 9.38 | 56.4 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 13.29 | 54.7 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 83.1 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 43.02 | 44.2 | Misses a Standard requirement |
Among the one-step solutions on LLMs with a conclusive Intento LQA result in Yiddish, each dearer one has fewer weighted errors. They run from $1.78 per million characters and 307.5 weighted errors per 1,000 words on xAI Grok 4.20 (non-reasoning) to $43.02 and 44.2 on Anthropic Claude Fable 5. The one-step solution on DeepL Next Gen, an LLM-based MT engine, costs $27.50 per million characters. Its translations have 83.1 weighted errors per 1,000 words, more than those from OpenAI GPT-5.6 Terra at $5.50 per million characters.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 71.9 |
| Gender neutrality | 2.7 |
| Tone of voice | 0 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
None of the one-step or two-step solutions (translation + post-edit) meet general translation quality in Yiddish. The seven one-step solutions with a conclusive result have 44.2 to 307.5 weighted errors per 1,000 words, and the two-step solutions 34.6 to 161.2, all over the bar. Three-step solutions (translation + LQA + repair) have 8.9 and 16.5 weighted errors per 1,000 words on Anthropic Claude Fable 5 and OpenAI GPT-5.5, under the bar, and 22.1 and 112.4 on OpenAI GPT-5.6 Terra and OpenAI GPT-4.1, over it.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Yiddish | Class |
|---|---|---|
| xAI Grok 4.20 (non-reasoning) | 307.5 | LLMs |
| OpenAI GPT-4.1 | 168.8 | LLMs |
| DeepL Next Gen | 83.1 | LLM-based MT engines |
| OpenAI GPT-5.6 Terra | 71.9 | LLMs |
| OpenAI GPT-5.6 Sol | 56.4 | LLMs |
| OpenAI GPT-5.5 | 54.7 | LLMs |
| Anthropic Claude Fable 5 | 44.2 | LLMs |
In one step, translations from the six shortlisted LLMs with a conclusive Intento LQA result have 44.2 to 307.5 weighted errors per 1,000 words in Yiddish. Those from DeepL Next Gen, an LLM-based MT engine, have 83.1 weighted errors per 1,000 words, 1.88 times as many as those from Anthropic Claude Fable 5. For the shortlisted MT engine, Google Cloud Advanced Translation API, the Intento LQA result is not conclusive. Of the six LLMs, Anthropic Claude Fable 5 and OpenAI GPT-5.6 Sol meet our language-support criteria.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | 0.699 | 9 | 2.43 | 9 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.695 | 11 | 2.56 | 10 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.691 | 12 | 2.67 | 14 | No | Not shortlisted |
| z.ai GLM-5 | 0.688 | 14 | 2.70 | 15 | No | Not shortlisted |
| MiniMax M2.5 | 0.686 | 15 | 3.32 | 19 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.686 | 16 | 2.97 | 17 | No | Not shortlisted |
| DeepSeek V3.2 | 0.686 | 17 | 2.61 | 11 | No | Not shortlisted |
| Google Gemma 4 | 0.660 | 18 | 3.86 | 23 | No | Not shortlisted |
| Mistral Large 3 | 0.648 | 20 | 3.55 | 20 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.637 | 22 | 3.55 | 21 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.619 | 23 | 5.70 | 24 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.598 | 24 | 2.73 | 16 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.588 | 25 | 5.84 | 25 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.577 | 26 | 6.38 | 26 | No | Not shortlisted |
| Cohere Command A Translate | 0.556 | 27 | 6.90 | 27 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.342 | 28 | 13.59 | 28 | No | Not shortlisted |
| OpenAI GPT-5.6 Luna | 0.745 | 1 | 1.83 | 1 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.721 | 2 | 2.07 | 5 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | 0.720 | 3 | 1.97 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.719 | 4 | 2.21 | 6 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.717 | 5 | 2.02 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | 0.716 | 6 | 1.92 | 2 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (non-reasoning) | 0.705 | 7 | 2.62 | 12 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 4.6 | 0.702 | 8 | 2.65 | 13 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-4.1 | 0.698 | 10 | 2.33 | 7 | Yes | Top 8 by MetricX |
| Google Gemini 3.5 Flash | 0.690 | 13 | 2.39 | 8 | Yes | Top 8 by MetricX |
| Google Cloud Advanced Translation API | 0.658 | 19 | 3.31 | 18 | Yes | Real-time place |
| Translated Lara 2 | 0.643 | 21 | 3.59 | 22 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Luna | Adjusted LLM | 1 | 0.745 | 1 | 1.83 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 2 | 0.721 | 5 | 2.07 | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | Adjusted LLM | 3 | 0.720 | 3 | 1.97 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 4 | 0.719 | 6 | 2.21 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 5 | 0.717 | 4 | 2.02 | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 6 | 0.716 | 2 | 1.92 | Top 8 by both metrics |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 7 | 0.705 | 12 | 2.62 | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 8 | 0.702 | 13 | 2.65 | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 9 | 0.699 | 9 | 2.43 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 10 | 0.698 | 7 | 2.33 | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 11 | 0.695 | 10 | 2.56 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 12 | 0.691 | 14 | 2.67 | Not shortlisted |
| Google Gemini 3.5 Flash | Adjusted LLM | 13 | 0.690 | 8 | 2.39 | Top 8 by MetricX |
| z.ai GLM-5 | Adjusted LLM | 14 | 0.688 | 15 | 2.70 | Not shortlisted |
| MiniMax M2.5 | Off-the-shelf LLM | 15 | 0.686 | 19 | 3.32 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 16 | 0.686 | 17 | 2.97 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 17 | 0.686 | 11 | 2.61 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 18 | 0.660 | 23 | 3.86 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 19 | 0.658 | 18 | 3.31 | Real-time place |
| Mistral Large 3 | Adjusted LLM | 20 | 0.648 | 20 | 3.55 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 21 | 0.643 | 22 | 3.59 | Real-time place |
| OpenAI gpt-oss-120b | Adjusted LLM | 22 | 0.637 | 21 | 3.55 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 23 | 0.619 | 24 | 5.70 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 24 | 0.598 | 16 | 2.73 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 25 | 0.588 | 25 | 5.84 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 26 | 0.577 | 26 | 6.38 | Not shortlisted |
| Cohere Command A Translate | Off-the-shelf LLM | 27 | 0.556 | 27 | 6.90 | Not shortlisted |
| AI21 Jamba Large 1.7 | Off-the-shelf LLM | 28 | 0.342 | 28 | 13.59 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.699 | — | 2.76 | 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 28 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 automatic metrics rank DeepL Next Gen 2nd on COMET-Kiwi and 5th on MetricX in Yiddish, and Anthropic Claude Fable 5 4th and 6th. Intento LQA orders them the other way: the one-step solution on DeepL Next Gen has 83.1 weighted errors per 1,000 words, and the one on Anthropic Claude Fable 5 44.2. The solution on OpenAI GPT-5.6 Luna, 1st on both metrics, has no conclusive Intento LQA result. Nearest the cut in the tail is Google Gemini 3.1 Pro Preview, 9th on both metrics.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.5, three-step | $653 |
| Standard | OpenAI GPT-5.5, three-step | $653 |
| Advanced | No solution | — |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.52 | LLMs |
| xAI Grok 4.20 (non-reasoning) | $1.78 | LLMs |
| OpenAI GPT-4.1 | $3.69 | LLMs |
| Google Gemini 3.5 Flash | $5.33 | LLMs |
| OpenAI GPT-5.6 Terra | $5.5 | LLMs |
| OpenAI GPT-5.6 Sol | $9.38 | LLMs |
| Anthropic Claude Sonnet 4.6 | $12.26 | LLMs |
| OpenAI GPT-5.5 | $13.29 | LLMs |
| Google Cloud Advanced Translation API | $20 | MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $43.02 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| z.ai GLM-5 | 0.927 | LLMs |
| Mistral Large 3 | 0.885 | LLMs |
| Mistral Medium 3.1 | 0.871 | LLMs |
| Anthropic Claude Fable 5 | 0.694 | LLMs |
| Anthropic Claude Opus 4.8 | 0.688 | LLMs |
| Anthropic Claude Sonnet 5 | 0.675 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.671 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.659 | LLMs |
| DeepSeek V3.2 | 0.637 | LLMs |
| MoonshotAI Kimi K2.5 | 0.629 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.619 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.617 | LLMs |
| Cohere Command A Translate | 0.603 | LLMs |
| MiniMax M2.5 | 0.598 | LLMs |
| AI21 Jamba Large 1.7 | 0.581 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.501 | LLMs |
| Google Gemma 4 | 0.498 | Open-weight LLMs |
| Google Gemini 3.5 Flash | 0.484 | LLMs |
| OpenAI GPT-5.6 Terra | 0.368 | LLMs |
| OpenAI GPT-5.5 | 0.365 | LLMs |
| OpenAI GPT-4.1 | 0.364 | LLMs |
| OpenAI GPT-5.6 Sol | 0.362 | LLMs |
| OpenAI GPT-5.6 Luna | 0.36 | LLMs |
Over the 23 models, a Yiddish translation has a median 0.62 tokens per character, 2.3 times as many as the English source. Translations from the five shortlisted OpenAI models have 0.36 to 0.368 tokens per character, and from Anthropic Claude Fable 5 0.694. With nearly the same tokens per character, the one-step solutions on the OpenAI models cost $0.52 to $13.29 per million characters. On Anthropic Claude Fable 5, the one-step solution costs $43.02 per million characters and the three-step one (translation + LQA + repair) $683.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Anthropic Claude Fable 5, three-step | 408 |
| Standard | Anthropic Claude Fable 5, three-step | 408 |
| Advanced | No solution | — |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| DeepL Next Gen | 1.7 | LLM-based MT engines |
| Translated Lara 2 | 3.3 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 3.3 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.7 | MT engines |
| OpenAI GPT-5.6 Luna | 18 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 42 | LLMs |
| OpenAI GPT-4.1 | 47 | LLMs |
| MiniMax M2.5 | 67 | LLMs |
| AI21 Jamba Large 1.7 | 68 | LLMs |
| Swiss AI Apertus-8B-2509 | 77 | Open-weight LLMs |
| OpenAI GPT-5.6 Terra | 78 | LLMs |
| Mistral Medium 3.1 | 80 | LLMs |
| Anthropic Claude Haiku 4.5 | 85 | LLMs |
| Google Gemini 3.5 Flash | 89 | LLMs |
| xAI Grok 4.20 (reasoning) | 101 | LLMs |
| Cohere Command A Translate | 105 | LLMs |
| DeepSeek V3.2 | 108 | LLMs |
| Mistral Large 3 | 129 | LLMs |
| Google Gemini 3.1 Pro Preview | 132 | LLMs |
| OpenAI GPT-5.6 Sol | 135 | LLMs |
| Anthropic Claude Sonnet 5 | 146 | LLMs |
| Anthropic Claude Sonnet 4.6 | 157 | LLMs |
| OpenAI GPT-5.5 | 167 | LLMs |
| Anthropic Claude Fable 5 | 174 | LLMs |
| Anthropic Claude Opus 4.8 | 192 | LLMs |
| Google Gemma 4 | 245 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 252 | LLMs |
| z.ai GLM-5 | 435 | 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 Yiddish.
The four real-time (<2 sec) models in Yiddish are the MT engine Google Cloud Advanced Translation API, the LLM-based MT engines DeepL Next Gen and Translated Lara 2 and the open-weight OpenAI gpt-oss-120b. Of the three on the shortlist, only the solution on DeepL Next Gen has a conclusive Intento LQA result, with 83.1 weighted errors per 1,000 words. None of the three meet formatting.
| 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 | three-step, with the gender-neutrality agent | 99.1% | ✗ | $1,116 | 440 |
| OpenAI GPT-5.5 | three-step, with the gender-neutrality agent | 100% | ✗ | $1,122 | 473 |
For Yiddish we check the formal register, with איר, and gender-neutral wording through neutral rephrasing and plurals, never a default gender. The other options in Yiddish are a masculine default, paired forms and neutral rephrasing alone. Translations from Translated Lara 2 and Google Cloud Advanced Translation API have 76.6% and 93.62% of the sentences in the formal register, and translations from every other shortlisted model 100%.
Over the 12 shortlisted solutions in Yiddish, the median share of sentences on which the three LQA judges agree is 90.7%. They agree most on the translations from Anthropic Claude Fable 5, on 94.9% of the sentences, and least on those from OpenAI GPT-5.6 Luna, on 79.1%.
Leave feedback