The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Lithuanian 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 Lithuanian. |
| Tone of voice | The formal register (Jūs) 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 Lithuanian, 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 Lithuanian, and neither leads on errors, cost and time at once. The three-step solution (translation + LQA + repair) on DeepSeek V3.2 has 16.9 total errors per 1,000 words, against 29.5 for the two-step solution (translation + post-edit) on Google Gemini 3.1 Pro Preview. It costs $548 against $4.91 per million characters and takes 257 against 99 seconds per 1,000 words. Optimizing for quality selects DeepSeek V3.2, and for cost or latency Google Gemini 3.1 Pro Preview.
A limit to real-time solutions (<2 sec) leaves three on the shortlist in Lithuanian: those on DeepL Next Gen and ModernMT Enterprise Edition miss the Basic requirements, and the one on Translated Lara 2 has no conclusive Intento LQA result.
On DeepSeek V3.2, the gender-neutrality agent raises the cost of the three-step solution (translation + LQA + repair) from $548 to $940 per million characters and its time from 257 to 565 seconds per 1,000 words. Its total errors rise from 16.9 to 24.7 per 1,000 words, and its weighted errors from 8.2 to 20.8, over the bar of general translation quality.
DeepSeek V3.2
three-step
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
DeepSeek V3.2
three-step
Google Gemini 3.1 Pro Preview
two-step
Google Gemini 3.1 Pro Preview
two-step
Google Gemini 3.1 Pro Preview
three-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
three-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
three-step with the gender-neutrality agent
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 | 28.5 | $4.91 | 99 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | two-step | Google Gemini 3.1 Pro Preview | 29.5 | $4.91 | 99 |
| Advanced | Tier 3: one solution | ☆ | with the gender-neutrality agent | Google Gemini 3.1 Pro Preview | 11.9 | $1,051 | 359 |
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, optimizing for quality selects the three-step solution (translation + LQA + repair) on DeepSeek V3.2, for cost the one-step solution on Google Gemini 3.1 Pro Preview and for latency the one on Anthropic Claude Fable 5. The Standard requirements remove the one on Anthropic Claude Fable 5, which gets 99.6% of the tags right, under the 100% bar. The one-step solution on Google Gemini 3.1 Pro Preview misses formatting, which the two-step solution (translation + post-edit) on the same model meets.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Mistral Large 3 | 0.82 | 62 | Misses a Standard requirement |
| DeepSeek V3.2 | 1.02 | 44.2 | Misses a Standard requirement |
| xAI Grok 4.20 (non-reasoning) | 1.16 | 63.6 | Misses a Standard requirement |
| z.ai GLM-5 | 2.17 | 52.7 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.65 | 35.9 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.91 | 18.8 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.61 | 27 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 32.6 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 30.97 | 19.4 | Misses a Standard requirement |
Among the one-step solutions in Lithuanian with a cost and a conclusive Intento LQA result, the five cheapest are on LLMs, at $0.82 to $3.65 per million characters and 35.9 to 63.6 weighted errors per 1,000 words. The one on Google Gemini 3.1 Pro Preview costs $4.91 per million characters and has 18.8 weighted errors per 1,000 words. The three dearer ones have more weighted errors: OpenAI GPT-5.5 at $11.61 per million characters and 27.0 per 1,000 words, DeepL Next Gen at $27.50 and 32.6, Anthropic Claude Fable 5 at $30.97 and 19.4.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 40 |
| Gender neutrality | 9.1 |
| Tone of voice | 0 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Five one-step solutions in Lithuanian with a conclusive Intento LQA result meet every Standard requirement but general translation quality, on OpenAI GPT-5.5, DeepSeek V3.2, z.ai GLM-5, Mistral Large 3 and xAI Grok 4.20 (non-reasoning). The closest, on OpenAI GPT-5.5, has 27.0 weighted errors per 1,000 words and 24.1 in two steps (translation + post-edit). Three-step solutions (translation + LQA + repair) cut the weighted errors on DeepSeek V3.2 from 44.2 to 8.2 per 1,000 words and on xAI Grok 4.20 (non-reasoning) from 63.6 to 34.6.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Lithuanian | Class |
|---|---|---|
| ModernMT Enterprise Edition | 86.7 | MT engines |
| xAI Grok 4.20 (non-reasoning) | 63.6 | LLMs |
| Mistral Large 3 | 62 | LLMs |
| z.ai GLM-5 | 52.7 | LLMs |
| DeepSeek V3.2 | 44.2 | LLMs |
| Google Gemini 3.5 Flash | 35.9 | LLMs |
| DeepL Next Gen | 32.6 | LLM-based MT engines |
| OpenAI GPT-5.5 | 27 | LLMs |
| Anthropic Claude Fable 5 | 19.4 | LLMs |
| Google Gemini 3.1 Pro Preview | 18.8 | LLMs |
No translation into Lithuanian: SpeakLeash Bielik-11B-v3.0-Instruct (wrong language), Tilde TildeOpen LLM engine (far too short), Widn AI Tower Anthill (far too short, wrong language), Widn AI Tower Vesuvius (far too short, wrong language).
In one step in Lithuanian, translations from the eight shortlisted LLMs with a conclusive Intento LQA result have 18.8 to 63.6 weighted errors per 1,000 words. Those from DeepL Next Gen, an LLM-based MT engine, have 32.6 weighted errors per 1,000 words. Those from ModernMT Enterprise Edition, the one shortlisted MT engine that is not LLM-based, have 86.7 weighted errors per 1,000 words and 27 major errors. The two LLMs that do not meet our language-support criteria are Mistral Large 3 and xAI Grok 4.20 (non-reasoning).
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| OpenAI GPT-4.1 | 0.826 | 10 | 1.94 | 10 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.825 | 12 | 1.99 | 12 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.823 | 14 | 2.01 | 13 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.823 | 15 | 2.15 | 21 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.821 | 16 | 2.14 | 20 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.821 | 17 | 2.02 | 14 | No | Not shortlisted |
| Microsoft Translator | 0.820 | 19 | 2.08 | 17 | No | Not shortlisted |
| Google Translation LLM | 0.814 | 20 | 2.12 | 18 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.813 | 21 | 2.33 | 23 | No | Not shortlisted |
| Cohere Command A Translate | 0.808 | 22 | 2.20 | 22 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.807 | 23 | 2.13 | 19 | No | Not shortlisted |
| MiniMax M2.5 | 0.806 | 24 | 2.62 | 26 | No | Not shortlisted |
| Google Gemma 4 | 0.804 | 25 | 2.45 | 24 | No | Not shortlisted |
| Amazon Translate | 0.804 | 26 | 2.94 | 30 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.793 | 27 | 2.69 | 27 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.786 | 28 | 2.47 | 25 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.784 | 29 | 2.73 | 28 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.782 | 30 | 2.74 | 29 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.773 | 31 | 3.44 | 31 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.744 | 32 | 3.53 | 32 | No | Not shortlisted |
| Cohere Tiny Aya | 0.739 | 33 | 3.83 | 33 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.722 | 34 | 4.52 | 34 | No | Not shortlisted |
| AppTek NMT | 0.613 | 35 | 6.62 | 35 | No | Not shortlisted |
| TartuNLP | 0.493 | 36 | 10.45 | 36 | No | Not shortlisted |
| DeepL Next Gen | 0.839 | 1 | 1.50 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.835 | 2 | 1.64 | 3 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.832 | 3 | 1.85 | 7 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.831 | 4 | 1.80 | 5 | Yes | Top 8 by both metrics |
| DeepSeek V3.2 | 0.831 | 5 | 1.98 | 11 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | 0.830 | 6 | 1.79 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.829 | 7 | 1.57 | 2 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.829 | 8 | 1.91 | 9 | Yes | Top 8 by COMET-Kiwi |
| xAI Grok 4.20 (non-reasoning) | 0.828 | 9 | 1.89 | 8 | Yes | Top 8 by MetricX |
| Mistral Large 3 | 0.826 | 11 | 1.81 | 6 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.824 | 13 | 2.05 | 15 | Yes | Real-time place |
| ModernMT Enterprise Edition | 0.821 | 18 | 2.06 | 16 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Off-the-shelf NMT | 1 | 0.839 | 1 | 1.50 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 2 | 0.835 | 3 | 1.64 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 3 | 0.832 | 7 | 1.85 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 4 | 0.831 | 5 | 1.80 | Top 8 by both metrics |
| DeepSeek V3.2 | Adjusted LLM | 5 | 0.831 | 11 | 1.98 | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | Adjusted LLM | 6 | 0.830 | 4 | 1.79 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 7 | 0.829 | 2 | 1.57 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 8 | 0.829 | 9 | 1.91 | Top 8 by COMET-Kiwi |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 9 | 0.828 | 8 | 1.89 | Top 8 by MetricX |
| OpenAI GPT-4.1 | Adjusted LLM | 10 | 0.826 | 10 | 1.94 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 11 | 0.826 | 6 | 1.81 | Top 8 by MetricX |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 12 | 0.825 | 12 | 1.99 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 13 | 0.824 | 15 | 2.05 | Real-time place |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.823 | 13 | 2.01 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 15 | 0.823 | 21 | 2.15 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 16 | 0.821 | 20 | 2.14 | Not shortlisted |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 17 | 0.821 | 14 | 2.02 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 18 | 0.821 | 16 | 2.06 | Real-time place |
| Microsoft Translator | Off-the-shelf NMT | 19 | 0.820 | 17 | 2.08 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 20 | 0.814 | 18 | 2.12 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 21 | 0.813 | 23 | 2.33 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 22 | 0.808 | 22 | 2.20 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 23 | 0.807 | 19 | 2.13 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 24 | 0.806 | 26 | 2.62 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 25 | 0.804 | 24 | 2.45 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 26 | 0.804 | 30 | 2.94 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 27 | 0.793 | 27 | 2.69 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 28 | 0.786 | 25 | 2.47 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 29 | 0.784 | 28 | 2.73 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 30 | 0.782 | 29 | 2.74 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 31 | 0.773 | 31 | 3.44 | Not shortlisted |
| AI21 Jamba Large 1.7 | Off-the-shelf LLM | 32 | 0.744 | 32 | 3.53 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 33 | 0.739 | 33 | 3.83 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 34 | 0.722 | 34 | 4.52 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 35 | 0.613 | 35 | 6.62 | Not shortlisted |
| TartuNLP | Off-the-shelf NMT | 36 | 0.493 | 36 | 10.45 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.829 | — | 1.65 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.829 | — | 1.88 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.827 | — | 1.85 | 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 36 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.
By Intento LQA, one-step translations from DeepL Next Gen, first on both metrics in Lithuanian, have 32.6 weighted errors per 1,000 words, and those from Google Gemini 3.1 Pro Preview, second on COMET-Kiwi and third on MetricX, 18.8. DeepSeek V3.2 is on the shortlist through COMET-Kiwi alone, fifth there and 11th on MetricX. One-step translations from it have 44.2 weighted errors per 1,000 words, and the three-step solution (translation + LQA + repair) on it 8.2. Nearest the cut in the tail is OpenAI GPT-4.1, 10th 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 | $4.91 |
| Standard | Google Gemini 3.1 Pro Preview, two-step | $4.91 |
| Advanced | Google Gemini 3.1 Pro Preview, three-step with the gender-neutrality agent | $1,051 |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.47 | LLMs |
| Mistral Large 3 | $0.82 | LLMs |
| DeepSeek V3.2 | $1.02 | LLMs |
| xAI Grok 4.20 (non-reasoning) | $1.16 | LLMs |
| z.ai GLM-5 | $2.17 | LLMs |
| Google Gemini 3.5 Flash | $3.65 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.91 | LLMs |
| OpenAI GPT-5.5 | $11.61 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $30.97 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Fable 5 | 0.52 | LLMs |
| Anthropic Claude Opus 4.8 | 0.52 | LLMs |
| Anthropic Claude Sonnet 5 | 0.517 | LLMs |
| Mistral Medium 3.1 | 0.451 | LLMs |
| Mistral Large 3 | 0.448 | LLMs |
| AI21 Jamba Large 1.7 | 0.433 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.413 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.409 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.405 | Open-weight LLMs |
| z.ai GLM-5 | 0.398 | LLMs |
| DeepSeek V3.2 | 0.393 | LLMs |
| Cohere Command A Translate | 0.391 | LLMs |
| MoonshotAI Kimi K2.5 | 0.383 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.346 | LLMs |
| Google Gemini 3.5 Flash | 0.344 | LLMs |
| Google Gemma 4 | 0.344 | Open-weight LLMs |
| OpenAI GPT-5.6 Terra | 0.34 | LLMs |
| OpenAI GPT-5.6 Luna | 0.338 | LLMs |
| OpenAI GPT-5.6 Sol | 0.338 | LLMs |
| OpenAI GPT-5.5 | 0.335 | LLMs |
| OpenAI GPT-4.1 | 0.334 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.325 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.321 | LLMs |
| MiniMax M2.5 | 0.313 | LLMs |
| Cohere Tiny Aya | 0.31 | LLMs |
Of the one-step solutions with a cost, the one on Anthropic Claude Fable 5 is the dearest, at $30.97 per million characters, with 0.52 tokens per character. Tokens alone do not set cost. The tokenizer of Mistral Large 3 produces 0.448 tokens per character, and the solution on it costs $0.82 per million characters, against $11.61 for OpenAI GPT-5.5 at 0.335. On DeepSeek V3.2 the cost follows the steps: $1.02 per million characters in one step, $11.12 in two (translation + post-edit) and $548 in three (translation + LQA + repair).
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Anthropic Claude Fable 5, one-step LLM adjusted for Standard requirements | 76 |
| Standard | Google Gemini 3.1 Pro Preview, two-step | 99 |
| Advanced | Google Gemini 3.1 Pro Preview, three-step with the gender-neutrality agent | 359 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.1 | MT engines |
| Microsoft Translator | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| Google Translation LLM | 0.6 | LLM-based MT engines |
| DeepL Next Gen | 0.8 | LLM-based MT engines |
| SYSTRAN PNMT API | 1 | MT engines |
| Amazon Translate | 1.2 | MT engines |
| Translated Lara 2 | 1.5 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.9 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.4 | MT engines |
| TartuNLP | 5.5 | MT engines |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| OpenAI GPT-4.1 | 17 | LLMs |
| Cohere Tiny Aya | 17 | LLMs |
| AppTek NMT | 19 | MT engines |
| Google Gemini 3.5 Flash | 20 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 21 | LLMs |
| OpenAI GPT-5.6 Terra | 29 | LLMs |
| DeepSeek V3.2 | 44 | LLMs |
| Mistral Medium 3.1 | 44 | LLMs |
| Anthropic Claude Haiku 4.5 | 45 | LLMs |
| Swiss AI Apertus-8B-2509 | 46 | Open-weight LLMs |
| MiniMax M2.5 | 46 | LLMs |
| Alibaba Machine Translation API | 48 | MT engines |
| xAI Grok 4.20 (reasoning) | 55 | LLMs |
| AI21 Jamba Large 1.7 | 58 | LLMs |
| Cohere Command A Translate | 59 | LLMs |
| OpenAI GPT-5.5 | 62 | LLMs |
| Mistral Large 3 | 62 | LLMs |
| Anthropic Claude Sonnet 5 | 69 | LLMs |
| OpenAI GPT-5.6 Sol | 75 | LLMs |
| Anthropic Claude Fable 5 | 76 | LLMs |
| Anthropic Claude Sonnet 4.6 | 82 | LLMs |
| Anthropic Claude Opus 4.8 | 90 | LLMs |
| Google Gemini 3.1 Pro Preview | 99 | LLMs |
| Alibaba Qwen3-235B-A22B | 136 | Open-weight LLMs |
| Google Gemma 4 | 166 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 307 | LLMs |
| z.ai GLM-5 | 453 | 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 Lithuanian.
The real-time models in Lithuanian are five MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. None of the three shortlisted real-time solutions, on DeepL Next Gen, ModernMT Enterprise Edition and Translated Lara 2, meet formatting. Translations from the first two have 32.6 and 86.7 weighted errors per 1,000 words. The fastest solution that meets every Basic requirement takes 76 seconds per 1,000 words, the fastest for the Standard ones 99 and the only one for the Advanced ones 359.
| 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 |
|---|---|---|---|---|---|
| DeepSeek V3.2 | three-step, with the gender-neutrality agent | 89.8% | ✗ | $940 | 565 |
| Google Gemini 3.1 Pro Preview | three-step, with the gender-neutrality agent | 100% | ✓ | $1,051 | 359 |
With the gender-neutrality agent, the three-step solution (translation + LQA + repair) on DeepSeek V3.2 misses general translation quality as well as gender neutrality in Lithuanian: its translations have 20.8 weighted errors per 1,000 words, against 8.2 without the agent.
Leave feedback