The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Nepali 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 Nepali. |
| Tone of voice | The formal register (तपाईं) 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 Nepali, 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 three-step solution (translation + LQA + repair) on Google Gemini 3.1 Pro Preview meets every Standard requirement in Nepali with 7.6 total errors per 1,000 words, at $563 per million characters and 269 seconds per 1,000 words. The one-step solution on OpenAI GPT-5.5 meets them with 19.8 total errors per 1,000 words, at $11.09 per million characters and 118 seconds per 1,000 words. Optimize for selects Google Gemini 3.1 Pro Preview for quality, and OpenAI GPT-5.5 for cost and latency.
Under the Standard requirements in Nepali, a limit of one or two steps leaves only the one-step solution on OpenAI GPT-5.5. So does a cost cap or a time limit at or above those of the one-step solution, $11.09 per million characters and 118 seconds per 1,000 words, and under those of the three-step one, $563 and 269 seconds. A limit from 7.6 to under 19.8 total errors per 1,000 words leaves only the three-step solution on Google Gemini 3.1 Pro Preview. Any cap or limit below these ranges leaves none.
For the Advanced requirements, the one-step solution on OpenAI GPT-5.5 meets gender neutrality without the gender-neutrality agent. So meeting them costs and takes the same as meeting the Standard ones: $11.09 per million characters and 118 seconds per 1,000 words. On Google Gemini 3.1 Pro Preview, the agent works with the three-step solution and raises its cost from $563 to $959 per million characters. With it, 99.1% of the sentences about people are gender-neutral, where we require 100%.
Google Gemini 3.1 Pro Preview
three-step
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
three-step
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
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 | ★ ☆ | one-step | OpenAI GPT-5.5 | 19.8 | $11.09 | 118 |
| Standard | Tier 1: simple solution + backup | ★ ☆ | one-step | OpenAI GPT-5.5 | 19.8 | $11.09 | 118 |
| Advanced | Tier 3: one solution | ★ | one-step | OpenAI GPT-5.5 | 19.8 | $11.09 | 118 |
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 Standard requirements keep both solutions that meet every Basic requirement in Nepali. The Advanced requirements, which add gender neutrality, remove the three-step solution (translation + LQA + repair) on Google Gemini 3.1 Pro Preview and keep the one-step solution on OpenAI GPT-5.5. So one step is enough at every level. Under the Basic and Standard requirements, Optimize for selects Google Gemini 3.1 Pro Preview for quality and OpenAI GPT-5.5 for cost and latency. Under the Advanced ones it selects OpenAI GPT-5.5 for all three.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Mistral Large 3 | 1.5 | 50.4 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.11 | 66.6 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.49 | 35.6 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.59 | 26.4 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 7.64 | 21.8 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 8.18 | 46.5 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.09 | 19.8 | Meets every Standard requirement |
| Google Translation LLM | 21.29 | 113.4 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 42.8 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 41.13 | 20.1 | Misses a Standard requirement |
The one-step solution on OpenAI GPT-5.5 costs $11.09 per million characters in Nepali and has 19.8 weighted errors per 1,000 words. None of the dearer one-step solutions with a conclusive Intento LQA result has fewer. The dearest of them, on Anthropic Claude Fable 5, costs $41.13 per million characters and has 20.1 weighted errors per 1,000 words. Below $11.09, the one-step LLM solutions cost $1.50 to $8.18 per million characters and have 21.8 to 66.6 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 39.2 |
| Tag handling | 0.3 |
| Gender neutrality | 0.2 |
| Tone of voice | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
General translation quality is the requirement most one-step solutions miss in Nepali. More steps clear it on one model. Translations from Google Gemini 3.1 Pro Preview have 26.4 weighted errors per 1,000 words in one step, 26.0 in two steps (translation + post-edit) and 7.3 in three (translation + LQA + repair). On Google Gemini 3.5 Flash and OpenAI GPT-4.1, three steps bring them to 24.4 and 30.7 weighted errors per 1,000 words, over the bar.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Nepali | Class |
|---|---|---|
| Google Translation LLM | 113.4 | LLM-based MT engines |
| OpenAI GPT-4.1 | 66.6 | LLMs |
| Mistral Large 3 | 50.4 | LLMs |
| Anthropic Claude Sonnet 5 | 46.5 | LLMs |
| DeepL Next Gen | 42.8 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 35.6 | LLMs |
| Google Gemini 3.1 Pro Preview | 26.4 | LLMs |
| OpenAI GPT-5.6 Sol | 21.8 | LLMs |
| Anthropic Claude Fable 5 | 20.1 | LLMs |
| OpenAI GPT-5.5 | 19.8 | LLMs |
Translations into Nepali from one-step solutions on the eight shortlisted LLMs have 19.8 to 66.6 weighted errors per 1,000 words. Those from OpenAI GPT-4.1 have the most, more than the 60 weighted errors per 1,000 words our language-support criteria allow. Translations from the LLM-based MT engines DeepL Next Gen and Google Translation LLM have 94.5% and 95.1% of the tags right, the lowest shares on the shortlist. None of the four open-weight LLMs we got translations from made the shortlist.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Opus 4.8 | 0.903 | 9 | 1.46 | 13 | No | Not shortlisted |
| z.ai GLM-5 | 0.901 | 10 | 1.41 | 9 | No | Not shortlisted |
| Google Gemma 4 | 0.900 | 12 | 1.41 | 10 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.898 | 13 | 1.43 | 11 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.897 | 15 | 1.46 | 14 | No | Not shortlisted |
| DeepSeek V3.2 | 0.894 | 16 | 1.54 | 16 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.892 | 17 | 1.75 | 19 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.890 | 18 | 1.48 | 15 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.878 | 20 | 1.67 | 17 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.877 | 21 | 1.77 | 20 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.871 | 22 | 2.13 | 27 | No | Not shortlisted |
| MiniMax M2.5 | 0.869 | 23 | 1.91 | 23 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.868 | 24 | 1.84 | 22 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.867 | 25 | 2.08 | 25 | No | Not shortlisted |
| Microsoft Translator | 0.861 | 26 | 2.12 | 26 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.852 | 27 | 2.62 | 29 | No | Not shortlisted |
| NiuTrans | 0.844 | 28 | 2.53 | 28 | No | Not shortlisted |
| Cohere Command A Translate | 0.827 | 29 | 1.83 | 21 | No | Not shortlisted |
| Cohere Tiny Aya | 0.818 | 30 | 2.77 | 30 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.818 | 31 | 2.96 | 31 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.770 | 32 | 3.95 | 32 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.476 | 33 | 11.57 | 33 | No | Not shortlisted |
| OpenAI GPT-5.6 Sol | 0.912 | 1 | 1.21 | 1 | Yes | Top 8 by both metrics |
| Mistral Large 3 | 0.908 | 2 | 1.33 | 4 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.907 | 3 | 1.41 | 8 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.905 | 4 | 1.32 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.905 | 5 | 1.46 | 12 | Yes | Top 8 by COMET-Kiwi |
| DeepL Next Gen | 0.904 | 6 | 1.36 | 5 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.904 | 7 | 1.37 | 6 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.903 | 8 | 1.39 | 7 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.900 | 11 | 1.31 | 2 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.898 | 14 | 1.68 | 18 | Yes | Real-time place |
| Google Translation LLM | 0.887 | 19 | 2.07 | 24 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Sol | Adjusted LLM | 1 | 0.912 | 1 | 1.21 | Top 8 by both metrics |
| Mistral Large 3 | Adjusted LLM | 2 | 0.908 | 4 | 1.33 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 3 | 0.907 | 8 | 1.41 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 4 | 0.905 | 3 | 1.32 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 5 | 0.905 | 12 | 1.46 | Top 8 by COMET-Kiwi |
| DeepL Next Gen | Off-the-shelf NMT | 6 | 0.904 | 5 | 1.36 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 7 | 0.904 | 6 | 1.37 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 8 | 0.903 | 7 | 1.39 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 9 | 0.903 | 13 | 1.46 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 10 | 0.901 | 9 | 1.41 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 11 | 0.900 | 2 | 1.31 | Top 8 by MetricX |
| Google Gemma 4 | Adjusted LLM | 12 | 0.900 | 10 | 1.41 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 13 | 0.898 | 11 | 1.43 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 14 | 0.898 | 18 | 1.68 | Real-time place |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 15 | 0.897 | 14 | 1.46 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 16 | 0.894 | 16 | 1.54 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 17 | 0.892 | 19 | 1.75 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 18 | 0.890 | 15 | 1.48 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 19 | 0.887 | 24 | 2.07 | Real-time place |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 20 | 0.878 | 17 | 1.67 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 21 | 0.877 | 20 | 1.77 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 22 | 0.871 | 27 | 2.13 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 23 | 0.869 | 23 | 1.91 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 24 | 0.868 | 22 | 1.84 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 25 | 0.867 | 25 | 2.08 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 26 | 0.861 | 26 | 2.12 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 27 | 0.852 | 29 | 2.62 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 28 | 0.844 | 28 | 2.53 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 29 | 0.827 | 21 | 1.83 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 30 | 0.818 | 30 | 2.77 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 31 | 0.818 | 31 | 2.96 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 32 | 0.770 | 32 | 3.95 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 33 | 0.476 | 33 | 11.57 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.909 | — | 1.27 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.908 | — | 1.37 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.885 | — | 1.52 | 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 33 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 and Intento LQA order the shortlist differently in Nepali. The one-step solution on OpenAI GPT-5.6 Sol, first on both metrics, has 21.8 weighted errors per 1,000 words and one critical error. The one on OpenAI GPT-5.5, seventh on COMET-Kiwi and sixth on MetricX, is the only one of the 10 one-step solutions with a conclusive Intento LQA result that meets general translation quality. Nearest the cut in the tail are Anthropic Claude Opus 4.8, ninth on COMET-Kiwi, and z.ai GLM-5, ninth on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $11.09 |
| Standard | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $11.09 |
| Advanced | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $11.09 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Mistral Large 3 | $1.5 | LLMs |
| OpenAI GPT-4.1 | $3.11 | LLMs |
| Google Gemini 3.5 Flash | $3.49 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.59 | LLMs |
| OpenAI GPT-5.6 Sol | $7.64 | LLMs |
| Anthropic Claude Sonnet 5 | $8.18 | LLMs |
| OpenAI GPT-5.5 | $11.09 | LLMs |
| Google Translation LLM | $21.29 | 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 | $41.13 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Mistral Medium 3.1 | 0.895 | LLMs |
| Mistral Large 3 | 0.894 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.806 | Open-weight LLMs |
| z.ai GLM-5 | 0.791 | LLMs |
| Anthropic Claude Opus 4.8 | 0.718 | LLMs |
| Anthropic Claude Sonnet 5 | 0.717 | LLMs |
| Anthropic Claude Fable 5 | 0.716 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.686 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.685 | LLMs |
| AI21 Jamba Large 1.7 | 0.663 | LLMs |
| Cohere Command A Translate | 0.644 | LLMs |
| DeepSeek V3.2 | 0.561 | LLMs |
| MoonshotAI Kimi K2.5 | 0.536 | LLMs |
| MiniMax M2.5 | 0.48 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.416 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.408 | LLMs |
| OpenAI GPT-4.1 | 0.326 | LLMs |
| OpenAI GPT-5.6 Sol | 0.322 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.32 | LLMs |
| OpenAI GPT-5.5 | 0.32 | LLMs |
| OpenAI GPT-5.6 Luna | 0.32 | LLMs |
| OpenAI GPT-5.6 Terra | 0.32 | LLMs |
| Google Gemini 3.5 Flash | 0.319 | LLMs |
| Google Gemma 4 | 0.314 | Open-weight LLMs |
| Cohere Tiny Aya | 0.303 | LLMs |
The tokenizers produce a median 0.54 tokens per character of Nepali, 2.0 times their count for English. Tokens alone do not set the one-step cost. The tokenizer of Mistral Large 3 produces 0.89 tokens per character, the most of the shortlisted LLMs, and the one-step solution on it costs $1.50 per million characters, the least. The tokenizers of Anthropic Claude Fable 5 and Anthropic Claude Sonnet 5 both produce 0.72 tokens per character, and the one-step solutions on them cost $41.13 and $8.18 per million characters.
| 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 | 118 |
| Standard | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 118 |
| Advanced | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 118 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| Google Cloud Advanced Translation API | 0.3 | MT engines |
| ModernMT Enterprise Edition | 0.4 | MT engines |
| Google Translation LLM | 0.6 | LLM-based MT engines |
| DeepL Next Gen | 0.8 | LLM-based MT engines |
| Translated Lara 2 | 2.1 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.3 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.1 | MT engines |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| Cohere Tiny Aya | 18 | LLMs |
| OpenAI GPT-4.1 | 22 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 23 | LLMs |
| Alibaba Machine Translation API | 29 | MT engines |
| DeepSeek V3.2 | 48 | LLMs |
| OpenAI GPT-5.6 Terra | 52 | LLMs |
| Google Gemini 3.5 Flash | 57 | LLMs |
| Anthropic Claude Haiku 4.5 | 58 | LLMs |
| NiuTrans | 58 | MT engines |
| xAI Grok 4.20 (reasoning) | 62 | LLMs |
| Mistral Medium 3.1 | 64 | LLMs |
| MiniMax M2.5 | 67 | LLMs |
| Swiss AI Apertus-8B-2509 | 69 | Open-weight LLMs |
| Cohere Command A Translate | 71 | LLMs |
| AI21 Jamba Large 1.7 | 78 | LLMs |
| Mistral Large 3 | 80 | LLMs |
| Anthropic Claude Sonnet 5 | 87 | LLMs |
| Google Gemini 3.1 Pro Preview | 88 | LLMs |
| Anthropic Claude Fable 5 | 96 | LLMs |
| Anthropic Claude Opus 4.8 | 108 | LLMs |
| Anthropic Claude Sonnet 4.6 | 116 | LLMs |
| OpenAI GPT-5.5 | 118 | LLMs |
| Google Gemma 4 | 122 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 138 | LLMs |
| MoonshotAI Kimi K2.5 | 222 | LLMs |
| Alibaba Qwen3-235B-A22B | 259 | Open-weight LLMs |
| z.ai GLM-5 | 307 | 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 Nepali.
Six of the seven real-time models in Nepali are MT engines, three of them LLM-based: DeepL Next Gen, Google Translation LLM and Translated Lara 2, all on the shortlist. The seventh, the open-weight OpenAI gpt-oss-120b, is the fastest LLM, at 2.3 seconds per 1,000 words. The solutions on DeepL Next Gen and Google Translation LLM meet tone of voice, but not general translation quality, tag handling or formatting. The one on Translated Lara 2 has no conclusive Intento LQA result.
| 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 |
|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | three-step, with the gender-neutrality agent | 99.1% | ✗ | $959 | 262 |
| OpenAI GPT-5.5 | one-step | 100% | ✓ | $11.09 | 118 |
For Nepali we check the formal register, with तपाईं, and gender-neutral wording through honorific-plural forms and neutral rephrasing, never a default gender. The one-step solution on OpenAI GPT-5.5 meets gender neutrality without the gender-neutrality agent. With the agent, the three-step solution (translation + LQA + repair) on Google Gemini 3.1 Pro Preview meets every other Advanced requirement, with 6.6 weighted errors per 1,000 words and no major error. We require 100% of the sentences about people to be gender-neutral.
For the 11 shortlisted solutions in Nepali, the three LQA judges agree on a median 94.9% of the sentences. Both pairs of automatic metrics agree above their thresholds, and neither result is borderline. With agreement this high, the Intento LQA results on this page carry weight.
Leave feedback