The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Slovenian 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 Slovenian. |
| Tone of voice | The formal register (vi (vikanje)) 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 Slovenian, 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 13 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.
Neither of the two one-step solutions that meet every Standard requirement in Slovenian leads on errors, cost and time at once. The one on Anthropic Claude Fable 5 has 17.2 total errors per 1,000 words, costs $28.46 per million characters and takes 63 seconds per 1,000 words. The one on OpenAI GPT-5.5 has 26.5 total errors per 1,000 words, costs $10.61 per million characters and takes 55 seconds per 1,000 words. Optimize for selects the first for quality and the second for cost and latency.
Under the Standard requirements, a cost cap of at most $10.60 per million characters or a time limit of at most 54 seconds per 1,000 words removes both solutions. Under the Basic ones, a cap of $4.59 per million characters or less leaves none. A real-time (<2 sec) limit removes every solution, since both real-time solutions on the shortlist, on DeepL Next Gen and Translated Lara 2, miss general translation quality. None of the one-step solutions meet every Advanced requirement.
With the gender-neutrality agent, the two-step solution (translation + post-edit) on OpenAI GPT-5.5 costs $49.13 per million characters, against $10.61 for the one-step solution on the same model. It takes 192 seconds per 1,000 words against 55, and its total errors fall from 26.5 to 18.1 per 1,000 words. On Anthropic Claude Fable 5, the three-step solution (translation + LQA + repair) with the agent takes 228 seconds per 1,000 words against 63, and its total errors fall from 17.2 to 8.9 per 1,000 words.
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 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
two-step with the gender-neutrality agent
OpenAI GPT-5.5
two-step with the gender-neutrality agent
OpenAI GPT-5.5
two-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 | ★×4 | one-step | Google Gemini 3.1 Pro Preview | 24.5 | $4.60 | 82 |
| Standard | Tier 1: simple solution + backup | ★★ | one-step | OpenAI GPT-5.5 | 26.5 | $10.61 | 55 |
| Advanced | Tier 3: one solution | ☆ | with the gender-neutrality agent | OpenAI GPT-5.5 | 18.1 | $49.13 | 192 |
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, Optimize for selects the solution on Anthropic Claude Fable 5 for quality, the one on Google Gemini 3.1 Pro Preview for cost and the one on OpenAI GPT-5.5 for latency. The Standard requirements remove the solutions on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.6 Sol, which miss formatting, so Optimize for selects the one on OpenAI GPT-5.5 for cost as well. The one solution that meets every Advanced requirement takes two steps with the gender-neutrality agent, so Optimize for selects it for all three.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| DeepSeek V3.2 | 0.97 | 32.3 | Misses a Standard requirement |
| z.ai GLM-5 | 2.01 | 43.8 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.95 | 42.2 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.44 | 26.4 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.6 | 17.5 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 5.7 | 44.8 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 6.27 | 29.3 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 7.33 | 18.8 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 10.61 | 18.8 | Meets every Standard requirement |
| Anthropic Claude Opus 4.8 | 14.33 | 20.4 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 25.4 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 20.1 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 28.46 | 10.9 | Meets every Standard requirement |
The one-step solution on Google Gemini 3.1 Pro Preview costs $4.60 per million characters and has 17.5 weighted errors per 1,000 words, but misses formatting. Of the one-step solutions that cost more, only the one on Anthropic Claude Fable 5, the dearest at $28.46 per million characters, has fewer, 10.9 weighted errors per 1,000 words. The solutions on OpenAI GPT-5.6 Sol and OpenAI GPT-5.5 have 18.8 weighted errors per 1,000 words each, at $7.33 and $10.61 per million characters.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 25.4 |
| Gender neutrality | 7.7 |
| Tone of voice | 0 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Two solutions miss general translation quality by a small margin in Slovenian. Translations from DeepL Next Gen have 20.1 weighted errors per 1,000 words and those from Anthropic Claude Opus 4.8 have 20.4. Tag handling has near misses too. Translations from Translated Lara 2 have 99.8% of the tags right, and those from Anthropic Claude Opus 4.8 and Anthropic Claude Sonnet 4.6 have 99.2%. All four solutions that meet general translation quality also meet tag handling.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Slovenian | Class |
|---|---|---|
| Anthropic Claude Sonnet 5 | 44.8 | LLMs |
| z.ai GLM-5 | 43.8 | LLMs |
| OpenAI GPT-4.1 | 42.2 | LLMs |
| DeepSeek V3.2 | 32.3 | LLMs |
| Anthropic Claude Sonnet 4.6 | 29.3 | LLMs |
| Google Gemini 3.5 Flash | 26.4 | LLMs |
| Translated Lara 2 | 25.4 | LLM-based MT engines |
| Anthropic Claude Opus 4.8 | 20.4 | LLMs |
| DeepL Next Gen | 20.1 | LLM-based MT engines |
| OpenAI GPT-5.5 | 18.8 | LLMs |
| OpenAI GPT-5.6 Sol | 18.8 | LLMs |
| Google Gemini 3.1 Pro Preview | 17.5 | LLMs |
| Anthropic Claude Fable 5 | 10.9 | LLMs |
No translation into Slovenian: 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).
The four solutions that meet general translation quality in Slovenian are all LLMs, with 10.9 to 18.8 weighted errors per 1,000 words. Translations from the two LLM-based MT engines on the shortlist, DeepL Next Gen and Translated Lara 2, have 20.1 and 25.4 weighted errors per 1,000 words, and those from the other LLMs 20.4 to 44.8. All 11 shortlisted LLMs support Slovenian by our measured criteria. What we got from SpeakLeash Bielik-11B-v3.0-Instruct, Widn AI Tower Anthill and Widn AI Tower Vesuvius was in the wrong language.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| xAI Grok 4.20 (reasoning) | 0.809 | 10 | 1.24 | 15 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.808 | 13 | 1.25 | 17 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.804 | 16 | 1.19 | 11 | No | Not shortlisted |
| Mistral Large 3 | 0.803 | 17 | 1.28 | 18 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.802 | 18 | 1.34 | 23 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.800 | 19 | 1.28 | 19 | No | Not shortlisted |
| Microsoft Translator | 0.800 | 20 | 1.25 | 16 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.797 | 21 | 1.23 | 14 | No | Not shortlisted |
| Google Translation LLM | 0.795 | 22 | 1.72 | 34 | No | Not shortlisted |
| Google Gemma 4 | 0.791 | 23 | 1.37 | 26 | No | Not shortlisted |
| MiniMax M2.5 | 0.790 | 24 | 1.35 | 24 | No | Not shortlisted |
| Cohere Command A Translate | 0.786 | 25 | 1.33 | 22 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.784 | 26 | 1.29 | 20 | No | Not shortlisted |
| Supertext LLM | 0.784 | 27 | 1.47 | 28 | No | Not shortlisted |
| NiuTrans | 0.782 | 28 | 1.47 | 27 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.781 | 29 | 1.30 | 21 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.781 | 30 | 2.00 | 36 | No | Not shortlisted |
| Cohere Tiny Aya | 0.778 | 31 | 1.63 | 31 | No | Not shortlisted |
| Amazon Translate | 0.777 | 32 | 1.65 | 33 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.765 | 33 | 1.64 | 32 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.764 | 34 | 1.63 | 30 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.745 | 35 | 1.89 | 35 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.744 | 36 | 1.51 | 29 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.660 | 37 | 3.15 | 37 | No | Not shortlisted |
| AppTek NMT | 0.616 | 38 | 5.79 | 38 | No | Not shortlisted |
| DeepL Next Gen | 0.820 | 1 | 1.09 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.819 | 2 | 1.11 | 2 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.816 | 3 | 1.12 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.815 | 4 | 1.17 | 8 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | 0.814 | 5 | 1.18 | 9 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | 0.814 | 6 | 1.20 | 12 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.5 | 0.813 | 7 | 1.21 | 13 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | 0.813 | 8 | 1.18 | 10 | Yes | Top 8 by COMET-Kiwi |
| Translated Lara 2 | 0.809 | 9 | 1.36 | 25 | Yes | Real-time place |
| Anthropic Claude Sonnet 4.6 | 0.808 | 11 | 1.16 | 6 | Yes | Top 8 by MetricX |
| OpenAI GPT-4.1 | 0.808 | 12 | 1.16 | 7 | Yes | Top 8 by MetricX |
| z.ai GLM-5 | 0.807 | 14 | 1.14 | 4 | Yes | Top 8 by MetricX |
| DeepSeek V3.2 | 0.807 | 15 | 1.15 | 5 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Off-the-shelf NMT | 1 | 0.820 | 1 | 1.09 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 2 | 0.819 | 2 | 1.11 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 3 | 0.816 | 3 | 1.12 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 4 | 0.815 | 8 | 1.17 | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 5 | 0.814 | 9 | 1.18 | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 6 | 0.814 | 12 | 1.20 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.5 | Adjusted LLM | 7 | 0.813 | 13 | 1.21 | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 8 | 0.813 | 10 | 1.18 | Top 8 by COMET-Kiwi |
| Translated Lara 2 | Adjusted NMT | 9 | 0.809 | 25 | 1.36 | Real-time place |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 10 | 0.809 | 15 | 1.24 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 11 | 0.808 | 6 | 1.16 | Top 8 by MetricX |
| OpenAI GPT-4.1 | Adjusted LLM | 12 | 0.808 | 7 | 1.16 | Top 8 by MetricX |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 13 | 0.808 | 17 | 1.25 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 14 | 0.807 | 4 | 1.14 | Top 8 by MetricX |
| DeepSeek V3.2 | Adjusted LLM | 15 | 0.807 | 5 | 1.15 | Top 8 by MetricX |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 16 | 0.804 | 11 | 1.19 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 17 | 0.803 | 18 | 1.28 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 18 | 0.802 | 23 | 1.34 | Not shortlisted |
| Tilde TildeOpen LLM engine | Adjusted NMT | 19 | 0.800 | 19 | 1.28 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 20 | 0.800 | 16 | 1.25 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 21 | 0.797 | 14 | 1.23 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 22 | 0.795 | 34 | 1.72 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 23 | 0.791 | 26 | 1.37 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 24 | 0.790 | 24 | 1.35 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 25 | 0.786 | 22 | 1.33 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 26 | 0.784 | 20 | 1.29 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 27 | 0.784 | 28 | 1.47 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 28 | 0.782 | 27 | 1.47 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 29 | 0.781 | 21 | 1.30 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 30 | 0.781 | 36 | 2.00 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 31 | 0.778 | 31 | 1.63 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 32 | 0.777 | 33 | 1.65 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 33 | 0.765 | 32 | 1.64 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 34 | 0.764 | 30 | 1.63 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 35 | 0.745 | 35 | 1.89 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 36 | 0.744 | 29 | 1.51 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 37 | 0.660 | 37 | 3.15 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 38 | 0.616 | 38 | 5.79 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.811 | — | 1.16 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.813 | — | 1.17 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.807 | — | 1.26 | 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 38 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.
OpenAI GPT-5.5 is on the shortlist through COMET-Kiwi alone, rank 7 against 13 on MetricX, and its solution meets every Standard requirement. None of the four solutions on the list through MetricX alone meet general translation quality. Translations from z.ai GLM-5, DeepSeek V3.2, Anthropic Claude Sonnet 4.6 and OpenAI GPT-4.1, ranks 4 to 7 on MetricX, have 29.3 to 43.8 weighted errors per 1,000 words. Nearest the cut in the tail are xAI Grok 4.20 (reasoning), rank 10 on COMET-Kiwi, and Swiss AI Apertus-8B-2509, rank 11 on MetricX.
| 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 | $4.60 |
| Standard | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $10.61 |
| Advanced | OpenAI GPT-5.5, two-step with the gender-neutrality agent | $49.13 |
| Model | US dollars per million characters | Class |
|---|---|---|
| DeepSeek V3.2 | $0.97 | LLMs |
| z.ai GLM-5 | $2.01 | LLMs |
| OpenAI GPT-4.1 | $2.95 | LLMs |
| Google Gemini 3.5 Flash | $3.44 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.6 | LLMs |
| Anthropic Claude Sonnet 5 | $5.7 | LLMs |
| Anthropic Claude Sonnet 4.6 | $6.27 | LLMs |
| OpenAI GPT-5.6 Sol | $7.33 | LLMs |
| OpenAI GPT-5.5 | $10.61 | LLMs |
| Anthropic Claude Opus 4.8 | $14.33 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $28.46 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.484 | LLMs |
| Anthropic Claude Fable 5 | 0.483 | LLMs |
| Anthropic Claude Sonnet 5 | 0.483 | LLMs |
| Mistral Medium 3.1 | 0.419 | LLMs |
| AI21 Jamba Large 1.7 | 0.415 | LLMs |
| Mistral Large 3 | 0.415 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.38 | Open-weight LLMs |
| DeepSeek V3.2 | 0.371 | LLMs |
| MoonshotAI Kimi K2.5 | 0.369 | LLMs |
| z.ai GLM-5 | 0.368 | LLMs |
| Cohere Command A Translate | 0.36 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.357 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.354 | LLMs |
| Google Gemini 3.5 Flash | 0.328 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.327 | LLMs |
| Google Gemma 4 | 0.326 | Open-weight LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.323 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.323 | LLMs |
| OpenAI GPT-4.1 | 0.306 | LLMs |
| OpenAI GPT-5.6 Terra | 0.306 | LLMs |
| OpenAI GPT-5.6 Luna | 0.305 | LLMs |
| OpenAI GPT-5.5 | 0.304 | LLMs |
| OpenAI GPT-5.6 Sol | 0.304 | LLMs |
| MiniMax M2.5 | 0.3 | LLMs |
| Cohere Tiny Aya | 0.291 | LLMs |
The tokenizers produce a median 0.35 tokens per character of Slovenian, 1.4 times the tokens per character of the English source. The tokenizers of Anthropic Claude Fable 5 and Anthropic Claude Sonnet 5 both produce 0.48 tokens per character, yet the one-step solutions on them cost $28.46 and $5.70 per million characters. The cheapest solution costs $4.60 per million characters under the Basic requirements, $10.61 under the Standard ones and $49.13 under the Advanced ones, where it takes two steps with the gender-neutrality agent.
| 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 | 55 |
| Standard | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 55 |
| Advanced | OpenAI GPT-5.5, two-step with the gender-neutrality agent | 192 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| Google Cloud Advanced Translation API | 0.4 | MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| Google Translation LLM | 0.9 | LLM-based MT engines |
| Translated Lara 2 | 1.2 | LLM-based MT engines |
| Amazon Translate | 1.2 | MT engines |
| SYSTRAN PNMT API | 1.7 | MT engines |
| OpenAI gpt-oss-120b | 2.3 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 2.9 | MT engines |
| Tilde TildeOpen LLM engine | 6.8 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| Google Gemini 3.5 Flash | 16 | LLMs |
| Supertext LLM | 16 | LLM-based MT engines |
| xAI Grok 4.20 (non-reasoning) | 16 | LLMs |
| Cohere Tiny Aya | 19 | LLMs |
| OpenAI GPT-4.1 | 25 | LLMs |
| Alibaba Machine Translation API | 26 | MT engines |
| OpenAI GPT-5.6 Terra | 29 | LLMs |
| xAI Grok 4.20 (reasoning) | 34 | LLMs |
| Anthropic Claude Haiku 4.5 | 35 | LLMs |
| MiniMax M2.5 | 49 | LLMs |
| Mistral Medium 3.1 | 49 | LLMs |
| Swiss AI Apertus-8B-2509 | 50 | Open-weight LLMs |
| Cohere Command A Translate | 53 | LLMs |
| OpenAI GPT-5.5 | 55 | LLMs |
| NiuTrans | 55 | MT engines |
| Mistral Large 3 | 60 | LLMs |
| Anthropic Claude Sonnet 5 | 62 | LLMs |
| Anthropic Claude Fable 5 | 63 | LLMs |
| AI21 Jamba Large 1.7 | 64 | LLMs |
| DeepSeek V3.2 | 65 | LLMs |
| Anthropic Claude Sonnet 4.6 | 68 | LLMs |
| OpenAI GPT-5.6 Sol | 72 | LLMs |
| Anthropic Claude Opus 4.8 | 78 | LLMs |
| Google Gemini 3.1 Pro Preview | 82 | LLMs |
| AppTek NMT | 96 | MT engines |
| Alibaba Qwen3-235B-A22B | 132 | Open-weight LLMs |
| Google Gemma 4 | 158 | Open-weight LLMs |
| z.ai GLM-5 | 214 | LLMs |
| MoonshotAI Kimi K2.5 | 261 | 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 Slovenian.
Of the nine real-time models in Slovenian, eight are MT engines or LLM-based MT engines, and the other is the open-weight OpenAI gpt-oss-120b, which is not on the shortlist. The two real-time solutions on the shortlist, on DeepL Next Gen and Translated Lara 2, miss general translation quality, with 20.1 and 25.4 weighted errors per 1,000 words. The solutions that meet every Standard requirement take 55 and 63 seconds per 1,000 words, and the one that meets every Advanced requirement takes 192 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 |
|---|---|---|---|---|---|
| Anthropic Claude Fable 5 | three-step, with the gender-neutrality agent | 97.3% | ✗ | $1,032 | 228 |
| OpenAI GPT-5.5 | two-step, with the gender-neutrality agent | 100% | ✓ | $49.13 | 192 |
For tone of voice in Slovenian we check the formal register, with vi (vikanje) as the form of address. For gender neutrality we check neutral rephrasing and impersonal forms. None of the 13 shortlisted one-step solutions meet gender neutrality, with a median of 7.7 gender-neutrality errors per 1,000 words. With the gender-neutrality agent, 100% of the sentences about people are gender-neutral in the translations from the two-step solution (translation + post-edit) on OpenAI GPT-5.5.
The three LQA judges agree on a median of 97.2% of the sentences of the shortlisted solutions in Slovenian, so the Intento LQA results on this page carry weight. With both pairs of automatic metrics in agreement, the ranking that built the shortlist carries weight too. The automatic scores only rank the models for the shortlist, and the weighted errors on this page come from Intento LQA.
Leave feedback