The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Latvian 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 Latvian. |
| 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 Latvian, 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. 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 four solutions that meet every Standard requirement in Latvian, the three-step solutions (translation + LQA + repair) on Anthropic Claude Fable 5 and OpenAI GPT-5.6 Terra have 10.8 and 16.0 total errors per 1,000 words. They cost $582 and $526 per million characters and take 230 and 175 seconds per 1,000 words. The two-step solution (translation + post-edit) on OpenAI GPT-5.5 has 21.3 total errors per 1,000 words, and optimizing for cost or latency selects it. Optimizing for quality selects Anthropic Claude Fable 5.
A cost cap of at most $525 per million characters or a time limit of at most 119 seconds per 1,000 words leaves only the two-step solution (translation + post-edit) on OpenAI GPT-5.5 among the solutions that meet every Standard requirement in Latvian. The cheapest of the three-step solutions (translation + LQA + repair) that meet them, on OpenAI GPT-5.6 Terra, costs $526 per million characters, and the fastest, on OpenAI GPT-4.1, takes 120 seconds per 1,000 words.
For the Advanced requirements, the gender-neutrality agent works with the three-step solutions (translation + LQA + repair) on Anthropic Claude Fable 5 and OpenAI GPT-5.6 Terra. On Anthropic Claude Fable 5 it raises the cost from $582 to $972 per million characters and the time from 230 to 274 seconds per 1,000 words. Total errors rise from 10.8 to 12.5 per 1,000 words, and weighted errors per 1,000 words from 7.3 to 12.5. On OpenAI GPT-5.6 Terra the cost goes from $526 to $885 per million characters.
Anthropic Claude Fable 5
three-step
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
three-step
OpenAI GPT-5.5
two-step
OpenAI GPT-5.5
two-step
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Anthropic Claude Fable 5
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 | OpenAI GPT-5.5 | 21.3 | $11.77 | 68 |
| Standard | Tier 2: two or more solutions, none simple | ☆×4 | two-step | OpenAI GPT-5.5 | 21.3 | $11.77 | 68 |
| Advanced | Tier 3: one solution | ☆ | with the gender-neutrality agent | Anthropic Claude Fable 5 | 12.5 | $972 | 274 |
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 simplest solution that meets every Basic requirement in Latvian is one-step, for the Standard requirements two-step (translation + post-edit), and for the Advanced ones three-step (translation + LQA + repair) with the gender-neutrality agent. The Standard requirements remove the one-step solution on OpenAI GPT-5.5 on formatting, and its two-step version takes its place. For the Basic and Standard requirements, optimizing for quality selects Anthropic Claude Fable 5, and optimizing for cost or latency OpenAI GPT-5.5.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-4.1 | 3.25 | 43.2 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.84 | 34.3 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 4.68 | 20.1 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.13 | 21.8 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 7.39 | 41.9 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.77 | 16.8 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 15.33 | 26.4 | Misses a Standard requirement |
| Google Translation LLM | 20.64 | 81.1 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 46.1 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 42.8 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 30.5 | 22.1 | Misses a Standard requirement |
Of the one-step solutions, the dearest, on Anthropic Claude Fable 5, has 22.1 weighted errors per 1,000 words, against 16.8 on OpenAI GPT-5.5. Of the solutions that meet every Standard requirement, the two-step solution (translation + post-edit) on OpenAI GPT-5.5 and the three-step solutions (translation + LQA + repair) on OpenAI GPT-5.6 Terra and Anthropic Claude Fable 5 are on the front of cost against weighted errors. The three-step solution on OpenAI GPT-4.1 is off it, at $575 per million characters.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 34.3 |
| Gender neutrality | 4.5 |
| Tone of voice | 0 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Of the one-step solutions in Latvian, only the one on OpenAI GPT-5.5 meets general translation quality, with 16.8 weighted errors per 1,000 words. It does not meet formatting, and its two-step version (translation + post-edit) meets both. Two-step solutions on Anthropic Claude Fable 5 and OpenAI GPT-5.6 Terra stay over the bar, at 20.1 and 22.4 weighted errors per 1,000 words. Three-step solutions (translation + LQA + repair) on these two models and OpenAI GPT-4.1 have 7.3 to 15.2 weighted errors per 1,000 words.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Latvian | Class |
|---|---|---|
| Google Translation LLM | 81.1 | LLM-based MT engines |
| Translated Lara 2 | 46.1 | LLM-based MT engines |
| OpenAI GPT-4.1 | 43.2 | LLMs |
| DeepL Next Gen | 42.8 | LLM-based MT engines |
| Anthropic Claude Sonnet 4.6 | 41.9 | LLMs |
| Google Gemini 3.5 Flash | 34.3 | LLMs |
| Anthropic Claude Opus 4.8 | 26.4 | LLMs |
| Anthropic Claude Fable 5 | 22.1 | LLMs |
| Google Gemini 3.1 Pro Preview | 21.8 | LLMs |
| OpenAI GPT-5.6 Terra | 20.1 | LLMs |
| OpenAI GPT-5.5 | 16.8 | LLMs |
No translation into Latvian: SpeakLeash Bielik-11B-v3.0-Instruct (wrong language), Widn AI Tower Anthill (far too short, wrong language), Widn AI Tower Vesuvius (far too short, wrong language).
In one step in Latvian, translations from the eight shortlisted LLMs have 16.8 to 43.2 weighted errors per 1,000 words, and those from the three shortlisted LLM-based MT engines, DeepL Next Gen, Translated Lara 2 and Google Translation LLM, 42.8, 46.1 and 81.1.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| z.ai GLM-5 | 0.802 | 11 | 2.71 | 10 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.801 | 12 | 2.97 | 17 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.801 | 13 | 2.72 | 11 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.800 | 14 | 2.82 | 12 | No | Not shortlisted |
| xAI Grok 4.20 (non-reasoning) | 0.798 | 15 | 2.67 | 9 | No | Not shortlisted |
| DeepSeek V3.2 | 0.794 | 17 | 2.91 | 15 | No | Not shortlisted |
| Microsoft Translator | 0.793 | 18 | 3.11 | 20 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.791 | 19 | 3.04 | 18 | No | Not shortlisted |
| MiniMax M2.5 | 0.789 | 20 | 3.26 | 24 | No | Not shortlisted |
| Mistral Large 3 | 0.787 | 21 | 3.08 | 19 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.783 | 22 | 3.23 | 22 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.780 | 23 | 3.13 | 21 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.778 | 24 | 3.23 | 23 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.773 | 25 | 3.52 | 27 | No | Not shortlisted |
| Google Gemma 4 | 0.769 | 26 | 3.58 | 28 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.767 | 27 | 3.36 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.763 | 28 | 3.49 | 26 | No | Not shortlisted |
| Amazon Translate | 0.755 | 29 | 4.30 | 32 | No | Not shortlisted |
| Cohere Command A Translate | 0.752 | 30 | 3.67 | 29 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.752 | 31 | 4.26 | 30 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.741 | 32 | 4.29 | 31 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.715 | 33 | 4.62 | 33 | No | Not shortlisted |
| Cohere Tiny Aya | 0.699 | 34 | 4.80 | 34 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.660 | 35 | 6.61 | 35 | No | Not shortlisted |
| AppTek NMT | 0.563 | 36 | 7.96 | 36 | No | Not shortlisted |
| TartuNLP | 0.479 | 37 | 11.36 | 37 | No | Not shortlisted |
| OpenAI GPT-5.6 Terra | 0.812 | 1 | 2.53 | 5 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.810 | 2 | 2.64 | 8 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.808 | 3 | 2.60 | 6 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.808 | 4 | 2.24 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.808 | 5 | 2.52 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.806 | 6 | 2.51 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | 0.805 | 7 | 2.82 | 13 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | 0.805 | 8 | 2.62 | 7 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.805 | 9 | 2.28 | 2 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.803 | 10 | 2.86 | 14 | Yes | Real-time place |
| Google Translation LLM | 0.794 | 16 | 2.94 | 16 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Terra | Adjusted LLM | 1 | 0.812 | 5 | 2.53 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 2 | 0.810 | 8 | 2.64 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 3 | 0.808 | 6 | 2.60 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 4 | 0.808 | 1 | 2.24 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 5 | 0.808 | 4 | 2.52 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 6 | 0.806 | 3 | 2.51 | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 7 | 0.805 | 13 | 2.82 | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 8 | 0.805 | 7 | 2.62 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 9 | 0.805 | 2 | 2.28 | Top 8 by MetricX |
| Translated Lara 2 | Adjusted NMT | 10 | 0.803 | 14 | 2.86 | Real-time place |
| z.ai GLM-5 | Adjusted LLM | 11 | 0.802 | 10 | 2.71 | Not shortlisted |
| Tilde TildeOpen LLM engine | Off-the-shelf NMT | 12 | 0.801 | 17 | 2.97 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 13 | 0.801 | 11 | 2.72 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 14 | 0.800 | 12 | 2.82 | Not shortlisted |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 15 | 0.798 | 9 | 2.67 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 16 | 0.794 | 16 | 2.94 | Real-time place |
| DeepSeek V3.2 | Adjusted LLM | 17 | 0.794 | 15 | 2.91 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 18 | 0.793 | 20 | 3.11 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 19 | 0.791 | 18 | 3.04 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 20 | 0.789 | 24 | 3.26 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 21 | 0.787 | 19 | 3.08 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 22 | 0.783 | 22 | 3.23 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.780 | 21 | 3.13 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 24 | 0.778 | 23 | 3.23 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 25 | 0.773 | 27 | 3.52 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 26 | 0.769 | 28 | 3.58 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 27 | 0.767 | 25 | 3.36 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 28 | 0.763 | 26 | 3.49 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 29 | 0.755 | 32 | 4.30 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 30 | 0.752 | 29 | 3.67 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 31 | 0.752 | 30 | 4.26 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 32 | 0.741 | 31 | 4.29 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 33 | 0.715 | 33 | 4.62 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 34 | 0.699 | 34 | 4.80 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 35 | 0.660 | 35 | 6.61 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 36 | 0.563 | 36 | 7.96 | Not shortlisted |
| TartuNLP | Off-the-shelf NMT | 37 | 0.479 | 37 | 11.36 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.805 | — | 2.61 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.808 | — | 2.56 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.792 | — | 3.02 | 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 37 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.
Google Gemini 3.5 Flash is on the shortlist through MetricX alone, second there and ninth on COMET-Kiwi. Anthropic Claude Sonnet 4.6 is on it through COMET-Kiwi alone, seventh there and 13th on MetricX. By Intento LQA, translations from them have 34.3 and 41.9 weighted errors per 1,000 words in one step. Translations from OpenAI GPT-5.6 Terra, first on COMET-Kiwi, have 20.1 weighted errors per 1,000 words. Nearest the cut in the tail are xAI Grok 4.20 (non-reasoning), 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 | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $11.77 |
| Standard | OpenAI GPT-5.5, two-step | $11.77 |
| Advanced | Anthropic Claude Fable 5, three-step with the gender-neutrality agent | $972 |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-4.1 | $3.25 | LLMs |
| Google Gemini 3.5 Flash | $3.84 | LLMs |
| OpenAI GPT-5.6 Terra | $4.68 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.13 | LLMs |
| Anthropic Claude Sonnet 4.6 | $7.39 | LLMs |
| OpenAI GPT-5.5 | $11.77 | LLMs |
| Anthropic Claude Opus 4.8 | $15.33 | LLMs |
| Google Translation LLM | $20.64 | 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 | $30.5 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.525 | LLMs |
| Anthropic Claude Sonnet 5 | 0.525 | LLMs |
| Anthropic Claude Fable 5 | 0.524 | LLMs |
| AI21 Jamba Large 1.7 | 0.476 | LLMs |
| Mistral Medium 3.1 | 0.474 | LLMs |
| Mistral Large 3 | 0.467 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.426 | Open-weight LLMs |
| Anthropic Claude Sonnet 4.6 | 0.424 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.423 | LLMs |
| z.ai GLM-5 | 0.416 | LLMs |
| DeepSeek V3.2 | 0.414 | LLMs |
| Cohere Command A Translate | 0.409 | LLMs |
| MoonshotAI Kimi K2.5 | 0.398 | LLMs |
| Google Gemma 4 | 0.368 | Open-weight LLMs |
| Google Gemini 3.1 Pro Preview | 0.367 | LLMs |
| Google Gemini 3.5 Flash | 0.362 | LLMs |
| OpenAI GPT-5.6 Luna | 0.345 | LLMs |
| OpenAI GPT-5.6 Terra | 0.345 | LLMs |
| MiniMax M2.5 | 0.343 | LLMs |
| OpenAI GPT-4.1 | 0.343 | LLMs |
| OpenAI GPT-5.5 | 0.343 | LLMs |
| OpenAI GPT-5.6 Sol | 0.343 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.341 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.341 | LLMs |
| Cohere Tiny Aya | 0.306 | LLMs |
On Anthropic Claude Fable 5 the cost follows the steps: $30.50 per million characters in one step, $67.85 in two, $582 in three and $972 with the gender-neutrality agent. The tokenizers of Anthropic Claude Fable 5 and Anthropic Claude Opus 4.8 produce 0.524 and 0.525 tokens per character of the Latvian translation. At about the same tokens per character, the one-step solution on Anthropic Claude Opus 4.8 costs $15.33 per million characters, against $30.50 on Anthropic Claude Fable 5.
| 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 | 68 |
| Standard | OpenAI GPT-5.5, two-step | 68 |
| Advanced | Anthropic Claude Fable 5, three-step with the gender-neutrality agent | 274 |
| 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 | 0.4 | MT engines |
| Google Translation LLM | 0.5 | LLM-based MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Amazon Translate | 1.2 | MT engines |
| SYSTRAN PNMT API | 1.5 | MT engines |
| Translated Lara 2 | 1.6 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.5 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 2.8 | MT engines |
| TartuNLP | 6.3 | MT engines |
| Tilde TildeOpen LLM engine | 6.5 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 16 | LLMs |
| Google Gemini 3.5 Flash | 18 | LLMs |
| Cohere Tiny Aya | 18 | LLMs |
| OpenAI GPT-4.1 | 22 | LLMs |
| OpenAI GPT-5.6 Terra | 34 | LLMs |
| xAI Grok 4.20 (reasoning) | 41 | LLMs |
| Alibaba Machine Translation API | 43 | MT engines |
| MiniMax M2.5 | 45 | LLMs |
| Anthropic Claude Haiku 4.5 | 47 | LLMs |
| Swiss AI Apertus-8B-2509 | 51 | Open-weight LLMs |
| Mistral Medium 3.1 | 58 | LLMs |
| Cohere Command A Translate | 61 | LLMs |
| AI21 Jamba Large 1.7 | 61 | LLMs |
| DeepSeek V3.2 | 62 | LLMs |
| OpenAI GPT-5.5 | 68 | LLMs |
| Anthropic Claude Sonnet 5 | 73 | LLMs |
| Anthropic Claude Fable 5 | 75 | LLMs |
| Mistral Large 3 | 80 | LLMs |
| OpenAI GPT-5.6 Sol | 80 | LLMs |
| Anthropic Claude Sonnet 4.6 | 87 | LLMs |
| AppTek NMT | 88 | MT engines |
| Anthropic Claude Opus 4.8 | 96 | LLMs |
| Google Gemini 3.1 Pro Preview | 98 | LLMs |
| Alibaba Qwen3-235B-A22B | 157 | Open-weight LLMs |
| Google Gemma 4 | 163 | Open-weight LLMs |
| z.ai GLM-5 | 251 | LLMs |
| MoonshotAI Kimi K2.5 | 252 | 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 Latvian.
The real-time models in Latvian are five MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. The three LLM-based MT engines, DeepL Next Gen, Translated Lara 2 and Google Translation LLM, are on the shortlist, and translations from all three meet tone of voice. None of the one-step solutions on them meet general translation quality, tag handling or 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 | 100% | ✓ | $972 | 274 |
| OpenAI GPT-5.6 Terra | three-step, with the gender-neutrality agent | 99.1% | ✗ | $885 | 175 |
With the gender-neutrality agent, the three-step solution (translation + LQA + repair) on OpenAI GPT-5.6 Terra misses general translation quality as well as gender neutrality in Latvian. Its translations have 17.5 weighted errors per 1,000 words, within the bar, but 4 major errors, one more than general translation quality allows.
For the 11 shortlisted solutions in Latvian, the median share of sentences on which the three LQA judges agree is 96.0%. Each pair of automatic metrics agrees above its threshold, and neither is borderline. So the Intento LQA error counts and the metric ranks behind the shortlist on this page rest on measures that agree in Latvian.
Leave feedback