The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Bulgarian 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 Bulgarian. |
| Tone of voice | The formal register (Вие) 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 Bulgarian, 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. 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.
Three one-step solutions meet every Standard requirement in Bulgarian. Those on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.6 Terra cost $4.69 and $4.67 per million characters. Translations from the first have 13.9 total errors per 1,000 words against 17.3, and take 82 seconds per 1,000 words against 26. So Optimize for selects Google Gemini 3.1 Pro Preview for quality and OpenAI GPT-5.6 Terra for cost and latency. The one on Anthropic Claude Fable 5 costs $26.14 per million characters, with 17.6 total errors per 1,000 words.
A real-time limit (<2 sec) leaves three shortlisted solutions in Bulgarian, on DeepL Next Gen, Google Translation LLM and OpenAI gpt-oss-120b, and all three miss general translation quality. The closest, on DeepL Next Gen, has 21.1 weighted errors per 1,000 words and four major errors, and it also misses tone of voice, tag handling and formatting.
For the Advanced requirements, the gender-neutrality agent raises the cost of the one-step solution on Google Gemini 3.1 Pro Preview from $4.69 to $4.94 per million characters and the time from 82 to 116 seconds per 1,000 words. Its translations have 19.4 total errors per 1,000 words instead of 13.9. On OpenAI GPT-5.6 Terra, the agent works with the three-step solution (translation + LQA + repair), at $977 per million characters and 159 seconds per 1,000 words, with 9.9 total errors per 1,000 words.
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
three-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
one-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
one-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 | OpenAI GPT-5.6 Terra | 17.3 | $4.67 | 26 |
| Standard | Tier 1: simple solution + backup | ★★★ | one-step | OpenAI GPT-5.6 Terra | 17.3 | $4.67 | 26 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | with the gender-neutrality agent | Google Gemini 3.1 Pro Preview | 19.4 | $4.94 | 116 |
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 remove the one-step solution on Anthropic Claude Opus 4.8, with 99.2% of the tags right, from the four that meet every Basic requirement. Both solutions that meet every Advanced requirement work with the gender-neutrality agent, in one step on Google Gemini 3.1 Pro Preview and three steps on OpenAI GPT-5.6 Terra. Optimize for selects Google Gemini 3.1 Pro Preview for quality and OpenAI GPT-5.6 Terra for cost and latency under the Basic and Standard requirements, and the reverse under the Advanced ones.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| MoonshotAI Kimi K2.5 | 1.44 | 36.3 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.22 | 22.4 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.47 | 23.4 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 4.67 | 14.2 | Meets every Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.69 | 11.2 | Meets every Standard requirement |
| Anthropic Claude Sonnet 5 | 5.25 | 28 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 7.24 | 27.4 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 13.11 | 12.9 | Misses a Standard requirement |
| Google Translation LLM | 21.15 | 55.4 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 26.14 | 14.5 | Meets every Standard requirement |
| DeepL Next Gen | 27.5 | 21.1 | Misses a Standard requirement |
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 22.9 |
| Gender neutrality | 3.1 |
| Tone of voice | 0.3 |
| Tag handling | 0.2 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Bulgarian | Class |
|---|---|---|
| OpenAI gpt-oss-120b | 63.3 | Open-weight LLMs |
| Google Translation LLM | 55.4 | LLM-based MT engines |
| MoonshotAI Kimi K2.5 | 36.3 | LLMs |
| Anthropic Claude Sonnet 5 | 28 | LLMs |
| Anthropic Claude Sonnet 4.6 | 27.4 | LLMs |
| Google Gemini 3.5 Flash | 23.4 | LLMs |
| OpenAI GPT-4.1 | 22.4 | LLMs |
| DeepL Next Gen | 21.1 | LLM-based MT engines |
| Anthropic Claude Fable 5 | 14.5 | LLMs |
| OpenAI GPT-5.6 Terra | 14.2 | LLMs |
| Anthropic Claude Opus 4.8 | 12.9 | LLMs |
| Google Gemini 3.1 Pro Preview | 11.2 | LLMs |
No translation into Bulgarian: SpeakLeash Bielik-11B-v3.0-Instruct (wrong language, wrong script), Widn AI Tower Anthill (far too short, wrong language, wrong script), Widn AI Tower Vesuvius (far too short, wrong language, wrong script).
Translations from the nine shortlisted LLMs that are not open-weight have 11.2 to 36.3 weighted errors per 1,000 words in Bulgarian, and the solutions on four of them meet general translation quality. Translations from DeepL Next Gen, the shortlisted LLM-based MT engine with the fewest, have 21.1 weighted errors per 1,000 words, 1.88 times as many as those from Google Gemini 3.1 Pro Preview. Translations from Google Translation LLM have 55.4 weighted errors per 1,000 words and 9 critical errors, the only ones on the shortlist.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.5 | 0.817 | 9 | 1.21 | 13 | No | Not shortlisted |
| z.ai GLM-5 | 0.816 | 10 | 1.18 | 12 | No | Not shortlisted |
| Mistral Large 3 | 0.813 | 13 | 1.15 | 10 | No | Not shortlisted |
| Google Gemma 4 | 0.812 | 14 | 1.27 | 14 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.811 | 15 | 1.35 | 19 | No | Not shortlisted |
| MiniMax M2.5 | 0.807 | 18 | 1.30 | 17 | No | Not shortlisted |
| DeepSeek V3.2 | 0.806 | 19 | 1.27 | 15 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.804 | 20 | 1.31 | 18 | No | Not shortlisted |
| Translated Lara 2 | 0.804 | 21 | 1.29 | 16 | No | Not shortlisted |
| Microsoft Translator | 0.803 | 22 | 1.37 | 20 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.803 | 23 | 1.37 | 21 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.800 | 24 | 1.52 | 26 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.799 | 25 | 1.41 | 24 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.796 | 26 | 1.39 | 23 | No | Not shortlisted |
| Cohere Command A Translate | 0.794 | 27 | 1.50 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.785 | 28 | 1.58 | 28 | No | Not shortlisted |
| Supertext LLM | 0.785 | 29 | 1.60 | 29 | No | Not shortlisted |
| Amazon Translate | 0.782 | 30 | 1.82 | 30 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.780 | 31 | 2.11 | 31 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.754 | 32 | 2.42 | 33 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.736 | 33 | 2.26 | 32 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.722 | 34 | 3.04 | 34 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.712 | 35 | 3.06 | 35 | No | Not shortlisted |
| Cohere Tiny Aya | 0.688 | 36 | 3.99 | 36 | No | Not shortlisted |
| AppTek NMT | 0.452 | 37 | 8.80 | 37 | No | Not shortlisted |
| DeepL Next Gen | 0.824 | 1 | 1.02 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | 0.822 | 2 | 1.07 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.822 | 3 | 1.10 | 4 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.821 | 4 | 1.13 | 6 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.821 | 5 | 1.16 | 11 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | 0.819 | 6 | 1.13 | 5 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.818 | 7 | 1.15 | 9 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 4.6 | 0.818 | 8 | 1.03 | 2 | Yes | Top 8 by both metrics |
| MoonshotAI Kimi K2.5 | 0.815 | 11 | 1.13 | 7 | Yes | Top 8 by MetricX |
| Anthropic Claude Sonnet 5 | 0.813 | 12 | 1.14 | 8 | Yes | Top 8 by MetricX |
| OpenAI gpt-oss-120b | 0.810 | 16 | 1.37 | 22 | Yes | Real-time place |
| Google Translation LLM | 0.807 | 17 | 1.53 | 27 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Off-the-shelf NMT | 1 | 0.824 | 1 | 1.02 | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | Adjusted LLM | 2 | 0.822 | 3 | 1.07 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 3 | 0.822 | 4 | 1.10 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 4 | 0.821 | 6 | 1.13 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 5 | 0.821 | 11 | 1.16 | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | Adjusted LLM | 6 | 0.819 | 5 | 1.13 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 7 | 0.818 | 9 | 1.15 | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 8 | 0.818 | 2 | 1.03 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 9 | 0.817 | 13 | 1.21 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 10 | 0.816 | 12 | 1.18 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 11 | 0.815 | 7 | 1.13 | Top 8 by MetricX |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 12 | 0.813 | 8 | 1.14 | Top 8 by MetricX |
| Mistral Large 3 | Adjusted LLM | 13 | 0.813 | 10 | 1.15 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 14 | 0.812 | 14 | 1.27 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 15 | 0.811 | 19 | 1.35 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 16 | 0.810 | 22 | 1.37 | Real-time place |
| Google Translation LLM | Adjusted NMT | 17 | 0.807 | 27 | 1.53 | Real-time place |
| MiniMax M2.5 | Adjusted LLM | 18 | 0.807 | 17 | 1.30 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 19 | 0.806 | 15 | 1.27 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 20 | 0.804 | 18 | 1.31 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 21 | 0.804 | 16 | 1.29 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 22 | 0.803 | 20 | 1.37 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.803 | 21 | 1.37 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 24 | 0.800 | 26 | 1.52 | Not shortlisted |
| Tilde TildeOpen LLM engine | Off-the-shelf NMT | 25 | 0.799 | 24 | 1.41 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 26 | 0.796 | 23 | 1.39 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 27 | 0.794 | 25 | 1.50 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 28 | 0.785 | 28 | 1.58 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 29 | 0.785 | 29 | 1.60 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 30 | 0.782 | 30 | 1.82 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 31 | 0.780 | 31 | 2.11 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 32 | 0.754 | 33 | 2.42 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 33 | 0.736 | 32 | 2.26 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 34 | 0.722 | 34 | 3.04 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 35 | 0.712 | 35 | 3.06 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 36 | 0.688 | 36 | 3.99 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 37 | 0.452 | 37 | 8.80 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.815 | — | 1.17 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.822 | — | 1.04 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.810 | — | 1.22 | 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 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.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | $4.67 |
| Standard | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | $4.67 |
| Advanced | Google Gemini 3.1 Pro Preview, one-step with the gender-neutrality agent | $4.94 |
| Model | US dollars per million characters | Class |
|---|---|---|
| MoonshotAI Kimi K2.5 | $1.44 | LLMs |
| OpenAI GPT-4.1 | $3.22 | LLMs |
| Google Gemini 3.5 Flash | $3.47 | LLMs |
| OpenAI GPT-5.6 Terra | $4.67 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.69 | LLMs |
| Anthropic Claude Sonnet 5 | $5.25 | LLMs |
| Anthropic Claude Sonnet 4.6 | $7.24 | LLMs |
| Anthropic Claude Opus 4.8 | $13.11 | LLMs |
| Google Translation LLM | $21.15 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $26.14 | LLMs |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| AI21 Jamba Large 1.7 | 0.55 | LLMs |
| Mistral Medium 3.1 | 0.422 | LLMs |
| Anthropic Claude Fable 5 | 0.417 | LLMs |
| Mistral Large 3 | 0.417 | LLMs |
| Anthropic Claude Opus 4.8 | 0.416 | LLMs |
| Anthropic Claude Sonnet 5 | 0.415 | LLMs |
| MoonshotAI Kimi K2.5 | 0.409 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.402 | Open-weight LLMs |
| Anthropic Claude Sonnet 4.6 | 0.387 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.384 | LLMs |
| DeepSeek V3.2 | 0.369 | LLMs |
| Cohere Command A Translate | 0.352 | LLMs |
| z.ai GLM-5 | 0.338 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.337 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.336 | LLMs |
| OpenAI GPT-5.5 | 0.325 | LLMs |
| OpenAI GPT-5.6 Terra | 0.325 | LLMs |
| OpenAI GPT-4.1 | 0.324 | LLMs |
| OpenAI GPT-5.6 Luna | 0.324 | LLMs |
| OpenAI GPT-5.6 Sol | 0.323 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.318 | LLMs |
| Google Gemini 3.5 Flash | 0.318 | LLMs |
| Google Gemma 4 | 0.316 | Open-weight LLMs |
| Cohere Tiny Aya | 0.312 | LLMs |
| MiniMax M2.5 | 0.312 | LLMs |
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | 26 |
| Standard | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | 26 |
| Advanced | Google Gemini 3.1 Pro Preview, one-step with the gender-neutrality agent | 116 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| Microsoft Translator | 0.3 | MT engines |
| ModernMT Enterprise Edition | 0.5 | MT engines |
| Google Translation LLM | 0.8 | LLM-based MT engines |
| DeepL Next Gen | 1 | LLM-based MT engines |
| Amazon Translate | 1.3 | MT engines |
| Translated Lara 2 | 1.5 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.7 | Open-weight LLMs |
| SYSTRAN PNMT API | 3.2 | MT engines |
| Yandex Cloud Translate v2 | 3.8 | MT engines |
| Tilde TildeOpen LLM engine | 7.2 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 15 | LLMs |
| Cohere Tiny Aya | 16 | LLMs |
| OpenAI GPT-5.6 Luna | 16 | LLMs |
| Supertext LLM | 17 | LLM-based MT engines |
| xAI Grok 4.20 (non-reasoning) | 20 | LLMs |
| OpenAI GPT-5.6 Terra | 26 | LLMs |
| OpenAI GPT-4.1 | 38 | LLMs |
| DeepSeek V3.2 | 40 | LLMs |
| Anthropic Claude Haiku 4.5 | 40 | LLMs |
| Mistral Medium 3.1 | 43 | LLMs |
| MiniMax M2.5 | 45 | LLMs |
| Swiss AI Apertus-8B-2509 | 51 | Open-weight LLMs |
| Cohere Command A Translate | 51 | LLMs |
| Alibaba Machine Translation API | 55 | MT engines |
| xAI Grok 4.20 (reasoning) | 57 | LLMs |
| Mistral Large 3 | 58 | LLMs |
| OpenAI GPT-5.6 Sol | 61 | LLMs |
| OpenAI GPT-5.5 | 62 | LLMs |
| Anthropic Claude Sonnet 5 | 66 | LLMs |
| Anthropic Claude Fable 5 | 68 | LLMs |
| AI21 Jamba Large 1.7 | 74 | LLMs |
| Anthropic Claude Sonnet 4.6 | 76 | LLMs |
| Google Gemini 3.1 Pro Preview | 82 | LLMs |
| Anthropic Claude Opus 4.8 | 87 | LLMs |
| AppTek NMT | 97 | MT engines |
| Alibaba Qwen3-235B-A22B | 153 | Open-weight LLMs |
| Google Gemma 4 | 159 | Open-weight LLMs |
| z.ai GLM-5 | 241 | LLMs |
| MoonshotAI Kimi K2.5 | 270 | 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 Bulgarian.
The nine real-time models in Bulgarian are five MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. DeepL Next Gen, Google Translation LLM and OpenAI gpt-oss-120b are on the shortlist, with 21.1, 55.4 and 63.3 weighted errors per 1,000 words, and the first two take 1.0 and 0.8 seconds per 1,000 words. The three solutions that meet every Standard requirement are on LLMs and take 26 to 82 seconds per 1,000 words. The fastest solution that meets every Advanced requirement takes 116 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 |
|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | one-step, with the gender-neutrality agent | 100% | ✓ | $4.94 | 116 |
| OpenAI GPT-5.6 Terra | three-step, with the gender-neutrality agent | 100% | ✓ | $977 | 159 |
For Bulgarian we check the formal register, with Вие, and gender-neutral wording through neutral rephrasing, plurals and impersonal forms, with no default gender. The other options in the language are feminine pairing or slash forms and the generic masculine. On Google Gemini 3.1 Pro Preview, the gender-neutrality agent raises weighted errors from 11.2 to 19.4 per 1,000 words and major errors from one to three. On OpenAI GPT-5.6 Terra, it works with the three-step solution (translation + LQA + repair).
The three LQA judges agree on a median of 96.6% of the sentences of the 12 shortlisted solutions in Bulgarian. Both pairs of automatic metrics agree above the threshold, and neither result is borderline. So the Intento LQA error counts and the shortlist ranks on this page are reliable.
Leave feedback