The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Uzbek 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 Uzbek. |
| 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 Uzbek, 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.
Of the two solutions that meet every Standard requirement in Uzbek, the three-step solution (translation + LQA + repair) on OpenAI GPT-5.5 has 5.6 total errors per 1,000 words and takes 223 seconds per 1,000 words. The two-step solution (translation + post-edit) on Google Gemini 3.1 Pro Preview has 17.5 total errors per 1,000 words and takes 263 seconds per 1,000 words, at $37.87 per million characters against $563. Optimizing for quality or latency selects the three-step solution, and optimizing for cost the two-step solution.
The two-step solution (translation + post-edit) on OpenAI GPT-5.5 takes 135 seconds per 1,000 words in Uzbek, less than the 223 and 263 of the two solutions that meet every Standard requirement. It misses only general translation quality, with 21.4 weighted errors per 1,000 words, just over the bar. So a time limit of 222 seconds per 1,000 words or less removes both solutions that meet the Standard requirements.
The Advanced requirements add gender neutrality to the Standard ones. In Uzbek, both solutions that meet every Standard requirement meet it without the gender-neutrality agent, with 100% of the sentences about people gender-neutral. So under the Advanced requirements the errors, cost and time are the same: 5.6 and 17.5 total errors per 1,000 words, $37.87 per million characters at the cheapest and 223 seconds per 1,000 words at the fastest.
OpenAI GPT-5.5
three-step
Google Gemini 3.1 Pro Preview
two-step
OpenAI GPT-5.5
three-step
OpenAI GPT-5.5
three-step
Google Gemini 3.1 Pro Preview
two-step
OpenAI GPT-5.5
three-step
OpenAI GPT-5.5
three-step
Google Gemini 3.1 Pro Preview
two-step
OpenAI GPT-5.5
three-step
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 | ☆☆ | two-step | Google Gemini 3.1 Pro Preview | 17.5 | $37.87 | 263 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | two-step | Google Gemini 3.1 Pro Preview | 17.5 | $37.87 | 263 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | two-step | Google Gemini 3.1 Pro Preview | 17.5 | $37.87 | 263 |
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 same two solutions meet every Basic, Standard and Advanced requirement in Uzbek, so moving up from the Basic requirements removes neither. Meeting any of the three sets of requirements takes at least two steps, and both solutions meet gender neutrality without the gender-neutrality agent. Under all three, optimizing for quality or latency selects the three-step solution (translation + LQA + repair) on OpenAI GPT-5.5. Optimizing for cost selects the two-step solution (translation + post-edit) on Google Gemini 3.1 Pro Preview.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-5.6 Luna | 0.49 | 32 | Misses a Standard requirement |
| xAI Grok 4.20 (non-reasoning) | 1.34 | 53.4 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 4.07 | 47.1 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.52 | 23.4 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 12.29 | 22.4 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 17.43 | 37.9 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 65.9 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 34.89 | 26.4 | Misses a Standard requirement |
Among the one-step solutions in Uzbek, a higher price does not always come with fewer errors. The dearest, on Anthropic Claude Fable 5 at $34.89 per million characters, has 26.4 weighted errors per 1,000 words. The one on Google Gemini 3.1 Pro Preview costs $5.52 per million characters and has 23.4 weighted errors per 1,000 words. The cheapest, on OpenAI GPT-5.6 Luna at $0.49 per million characters, has 32.0 weighted errors per 1,000 words, and the one on DeepL Next Gen, at $27.50, has 65.9.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 35 |
| 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.
With two steps (translation + post-edit), the solution on Google Gemini 3.1 Pro Preview has 17.5 weighted errors per 1,000 words in Uzbek, under the bar, and the one on OpenAI GPT-5.5 has 21.4, just over it. Of the one-step solutions, the four on OpenAI GPT-5.6 Luna, Anthropic Claude Opus 4.8, Google Gemini 3.5 Flash and xAI Grok 4.20 (non-reasoning) also miss tag handling, with 98.4% to 99.2% of the tags right, and the one on DeepL Next Gen has 94.9%.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Uzbek | Class |
|---|---|---|
| DeepL Next Gen | 65.9 | LLM-based MT engines |
| xAI Grok 4.20 (non-reasoning) | 53.4 | LLMs |
| Google Gemini 3.5 Flash | 47.1 | LLMs |
| Anthropic Claude Opus 4.8 | 37.9 | LLMs |
| OpenAI GPT-5.6 Luna | 32 | LLMs |
| Anthropic Claude Fable 5 | 26.4 | LLMs |
| Google Gemini 3.1 Pro Preview | 23.4 | LLMs |
| OpenAI GPT-5.5 | 22.4 | LLMs |
Translations from DeepL Next Gen, an LLM-based MT engine, have 65.9 weighted errors per 1,000 words and 15 major errors in the four test texts, the most of the eight one-step solutions with a conclusive result in Uzbek. Those from the other seven, all LLMs, have 22.4 to 53.4 weighted errors per 1,000 words.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| z.ai GLM-5 | 0.775 | 11 | 1.98 | 13 | No | Not shortlisted |
| Google Gemma 4 | 0.774 | 12 | 1.87 | 10 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.773 | 13 | 1.83 | 9 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.770 | 14 | 2.02 | 14 | No | Not shortlisted |
| Cohere Command A Translate | 0.768 | 15 | 2.17 | 17 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.767 | 16 | 2.16 | 16 | No | Not shortlisted |
| Mistral Large 3 | 0.764 | 17 | 1.97 | 12 | No | Not shortlisted |
| DeepSeek V3.2 | 0.764 | 18 | 2.18 | 18 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.764 | 19 | 2.18 | 19 | No | Not shortlisted |
| Google Translation LLM | 0.761 | 21 | 2.38 | 22 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.759 | 22 | 2.38 | 21 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.743 | 23 | 2.53 | 23 | No | Not shortlisted |
| MiniMax M2.5 | 0.743 | 24 | 2.62 | 24 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.734 | 25 | 3.17 | 27 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.734 | 26 | 2.99 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.725 | 27 | 3.00 | 26 | No | Not shortlisted |
| Amazon Translate | 0.720 | 28 | 3.29 | 28 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.715 | 29 | 3.56 | 29 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.482 | 30 | 9.25 | 30 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.181 | 31 | 18.32 | 31 | No | Not shortlisted |
| OpenAI GPT-5.5 | 0.784 | 1 | 1.78 | 6 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.782 | 2 | 1.63 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.782 | 3 | 1.63 | 2 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.782 | 4 | 1.65 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.780 | 5 | 1.82 | 7 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.780 | 6 | 1.76 | 5 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.778 | 7 | 1.82 | 8 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (non-reasoning) | 0.777 | 8 | 1.74 | 4 | Yes | Top 8 by both metrics |
| Translated Lara 2 | 0.775 | 9 | 2.13 | 15 | Yes | Real-time place |
| DeepL Next Gen | 0.775 | 10 | 1.93 | 11 | Yes | Real-time place |
| Microsoft Translator | 0.764 | 20 | 2.24 | 20 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.5 | Adjusted LLM | 1 | 0.784 | 6 | 1.78 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 2 | 0.782 | 1 | 1.63 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 3 | 0.782 | 2 | 1.63 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 4 | 0.782 | 3 | 1.65 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 5 | 0.780 | 7 | 1.82 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 6 | 0.780 | 5 | 1.76 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 7 | 0.778 | 8 | 1.82 | Top 8 by both metrics |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 8 | 0.777 | 4 | 1.74 | Top 8 by both metrics |
| Translated Lara 2 | Adjusted NMT | 9 | 0.775 | 15 | 2.13 | Real-time place |
| DeepL Next Gen | Off-the-shelf NMT | 10 | 0.775 | 11 | 1.93 | Real-time place |
| z.ai GLM-5 | Adjusted LLM | 11 | 0.775 | 13 | 1.98 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 12 | 0.774 | 10 | 1.87 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 13 | 0.773 | 9 | 1.83 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 14 | 0.770 | 14 | 2.02 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 15 | 0.768 | 17 | 2.17 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 16 | 0.767 | 16 | 2.16 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 17 | 0.764 | 12 | 1.97 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 18 | 0.764 | 18 | 2.18 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 19 | 0.764 | 19 | 2.18 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 20 | 0.764 | 20 | 2.24 | Real-time place |
| Google Translation LLM | Adjusted NMT | 21 | 0.761 | 22 | 2.38 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 22 | 0.759 | 21 | 2.38 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.743 | 23 | 2.53 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 24 | 0.743 | 24 | 2.62 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 25 | 0.734 | 27 | 3.17 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 26 | 0.734 | 25 | 2.99 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 27 | 0.725 | 26 | 3.00 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 28 | 0.720 | 28 | 3.29 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 29 | 0.715 | 29 | 3.56 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 30 | 0.482 | 30 | 9.25 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 31 | 0.181 | 31 | 18.32 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.779 | — | 1.67 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.778 | — | 1.73 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.776 | — | 1.86 | 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 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.
COMET-Kiwi and MetricX put the same eight LLMs in their top 8 in Uzbek, in different orders. MetricX ranks OpenAI GPT-5.5 sixth and xAI Grok 4.20 (non-reasoning) fourth. Yet by Intento LQA, the one-step solutions on these two models have 22.4 and 53.4 weighted errors per 1,000 words, the fewest and the most of the seven shortlisted LLMs with a conclusive result. Nearest the cut in the tail are Anthropic Claude Sonnet 4.6, ninth on MetricX, and z.ai GLM-5, 11th on COMET-Kiwi.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, two-step | $37.87 |
| Standard | Google Gemini 3.1 Pro Preview, two-step | $37.87 |
| Advanced | Google Gemini 3.1 Pro Preview, two-step | $37.87 |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.49 | LLMs |
| xAI Grok 4.20 (non-reasoning) | $1.34 | LLMs |
| Google Gemini 3.5 Flash | $4.07 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.52 | LLMs |
| Anthropic Claude Sonnet 5 | $6.98 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $12.29 | LLMs |
| Anthropic Claude Opus 4.8 | $17.43 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $34.89 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Sonnet 4.6 | 0.633 | LLMs |
| Anthropic Claude Sonnet 5 | 0.543 | LLMs |
| Anthropic Claude Fable 5 | 0.539 | LLMs |
| Anthropic Claude Opus 4.8 | 0.538 | LLMs |
| AI21 Jamba Large 1.7 | 0.496 | LLMs |
| Mistral Medium 3.1 | 0.435 | LLMs |
| Mistral Large 3 | 0.431 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.414 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.389 | Open-weight LLMs |
| DeepSeek V3.2 | 0.386 | LLMs |
| z.ai GLM-5 | 0.381 | LLMs |
| Cohere Command A Translate | 0.378 | LLMs |
| MoonshotAI Kimi K2.5 | 0.365 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.363 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.362 | LLMs |
| MiniMax M2.5 | 0.35 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.349 | LLMs |
| Google Gemini 3.5 Flash | 0.349 | LLMs |
| Google Gemma 4 | 0.347 | Open-weight LLMs |
| OpenAI GPT-4.1 | 0.325 | LLMs |
| OpenAI GPT-5.5 | 0.321 | LLMs |
| OpenAI GPT-5.6 Terra | 0.321 | LLMs |
| OpenAI GPT-5.6 Luna | 0.32 | LLMs |
| OpenAI GPT-5.6 Sol | 0.319 | LLMs |
The tokenizers of the three shortlisted Anthropic Claude models produce 0.538 to 0.543 tokens per character of the translation, against 0.321 for OpenAI GPT-5.5. Yet the one-step solution on Anthropic Claude Sonnet 5 costs $6.98 per million characters, against $12.29 on OpenAI GPT-5.5. On OpenAI GPT-5.5, the two-step solution (translation + post-edit) costs $44.85 per million characters and the three-step one (translation + LQA + repair) $563.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.5, three-step | 223 |
| Standard | OpenAI GPT-5.5, three-step | 223 |
| Advanced | OpenAI GPT-5.5, three-step | 223 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.1 | MT engines |
| Google Translation LLM | 0.6 | LLM-based MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| Microsoft Translator | 0.7 | MT engines |
| Amazon Translate | 1.1 | MT engines |
| Translated Lara 2 | 2 | LLM-based MT engines |
| Yandex Cloud Translate v2 | 2.7 | MT engines |
| OpenAI gpt-oss-120b | 3 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | 16 | LLMs |
| Google Gemini 3.5 Flash | 19 | LLMs |
| OpenAI GPT-4.1 | 22 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 23 | LLMs |
| Alibaba Machine Translation API | 29 | MT engines |
| OpenAI GPT-5.6 Terra | 31 | LLMs |
| Swiss AI Apertus-8B-2509 | 45 | Open-weight LLMs |
| DeepSeek V3.2 | 49 | LLMs |
| Anthropic Claude Haiku 4.5 | 49 | LLMs |
| Mistral Medium 3.1 | 50 | LLMs |
| OpenAI GPT-5.6 Sol | 62 | LLMs |
| xAI Grok 4.20 (reasoning) | 62 | LLMs |
| OpenAI GPT-5.5 | 63 | LLMs |
| Cohere Command A Translate | 64 | LLMs |
| AI21 Jamba Large 1.7 | 66 | LLMs |
| MiniMax M2.5 | 70 | LLMs |
| Mistral Large 3 | 74 | LLMs |
| Anthropic Claude Sonnet 5 | 85 | LLMs |
| Anthropic Claude Fable 5 | 87 | LLMs |
| Anthropic Claude Sonnet 4.6 | 90 | LLMs |
| Anthropic Claude Opus 4.8 | 109 | LLMs |
| Alibaba Qwen3-235B-A22B | 135 | Open-weight LLMs |
| Google Gemini 3.1 Pro Preview | 136 | LLMs |
| Google Gemma 4 | 178 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 202 | LLMs |
| z.ai GLM-5 | 262 | 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 Uzbek.
The real-time models in Uzbek are three MT engines, three LLM-based MT engines 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. Its translations meet tone of voice and have 94.9% of the tags right. The fastest shortlisted LLM solution, on OpenAI GPT-5.6 Luna, takes 16 seconds per 1,000 words and has 32.0 weighted errors per 1,000 words. The two solutions that meet every Advanced requirement take 223 and 263 seconds per 1,000 words.
| 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 | 100% | ✓ | $563 | 223 |
| Google Gemini 3.1 Pro Preview | two-step | 100% | ✓ | $37.87 | 263 |
For Uzbek we check the formal register, with Siz. Uzbek has no grammatical gender for persons, so for gender neutrality we check that a translation avoids gendered terms, does not add a gender the source does not specify and does not use a default gender. Translations from the two best solutions have 100% of the sentences about people gender-neutral without the gender-neutrality agent, so in Uzbek the agent is not needed for them.
For the 11 shortlisted solutions in Uzbek, the median share of sentences on which the three LQA judges agree is 93.2%. Both pairs of automatic metrics agree above their thresholds, and neither result is borderline. So the Intento LQA error counts and the automatic scores on this page carry weight in Uzbek.
Leave feedback