The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Galician 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 Galician. |
| Tone of voice | The formal register (vostede) 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 Galician, 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.
Of the three solutions that meet every Standard requirement in Galician, the one on OpenAI GPT-5.6 Sol has the fewest total errors, 15.2 per 1,000 words, and the shortest time, 48 seconds per 1,000 words, at $6.70 per million characters. The one on Google Gemini 3.5 Flash costs the least, $3.01 per million characters, with 29.7 total errors per 1,000 words, and the one on Anthropic Claude Fable 5 costs the most, $27.02, with 20.6. So Optimize for selects OpenAI GPT-5.6 Sol for quality and latency and Google Gemini 3.5 Flash for cost.
For the Standard requirements, a cost cap of $6.69 per million characters or less leaves only the solution on Google Gemini 3.5 Flash, and a cap of $3.00 or less leaves none. A time limit of 52 seconds per 1,000 words or less leaves only the one on OpenAI GPT-5.6 Sol, and a limit of 47 seconds or less leaves none. None of the three real-time (<2 sec) solutions on the shortlist, on Microsoft Translator, DeepL Next Gen and Translated Lara 2, meet general translation quality or formatting.
For the Advanced requirements, the solution on Anthropic Claude Fable 5 is a two-step solution (translation + post-edit) with the gender-neutrality agent. Against its one-step solution without the agent, it costs $53.46 instead of $27.02 per million characters, takes 102 instead of 53 seconds per 1,000 words and has 14.5 instead of 20.6 total errors per 1,000 words, from the post-edit step and the agent together. With the agent, the one on OpenAI GPT-5.6 Sol leads on errors, cost and time, and Optimize for selects it for all three.
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
Google Gemini 3.5 Flash
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
Google Gemini 3.5 Flash
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step with the gender-neutrality agent
OpenAI GPT-5.6 Sol
one-step with the gender-neutrality agent
OpenAI GPT-5.6 Sol
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 | Google Gemini 3.5 Flash | 29.7 | $3.01 | 60 |
| Standard | Tier 1: simple solution + backup | ★★★ | one-step | Google Gemini 3.5 Flash | 29.7 | $3.01 | 60 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | with the gender-neutrality agent | OpenAI GPT-5.6 Sol | 13.2 | $7.20 | 35 |
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.
From the Basic to the Standard requirements, tag handling removes the solution on Anthropic Claude Opus 4.8. The Advanced requirements keep the two best, on OpenAI GPT-5.6 Sol as a one-step solution and on Anthropic Claude Fable 5 as a two-step solution (translation + post-edit), both with the gender-neutrality agent. Under the Basic and Standard requirements, Optimize for selects OpenAI GPT-5.6 Sol for quality and latency and Google Gemini 3.5 Flash for cost. Under the Advanced ones, it selects OpenAI GPT-5.6 Sol for all three.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| xAI Grok 4.20 (reasoning) | 1.01 | 33.3 | Misses a Standard requirement |
| MoonshotAI Kimi K2.5 | 1.13 | 32.3 | Misses a Standard requirement |
| z.ai GLM-5 | 1.7 | 27.4 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.72 | 25.7 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.01 | 19.1 | Meets every Standard requirement |
| OpenAI GPT-5.6 Sol | 6.7 | 13.2 | Meets every Standard requirement |
| Microsoft Translator | 10 | 48.5 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 13.64 | 12.2 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 32.3 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 27.02 | 11.5 | Meets every Standard requirement |
| DeepL Next Gen | 27.5 | 38.6 | Misses a Standard requirement |
All three solutions that meet every Standard requirement are on the front of cost and errors, and all three are on LLMs. None of the solutions above $6.70 per million characters have fewer than 11.5 weighted errors per 1,000 words, against 13.2 for OpenAI GPT-5.6 Sol at $6.70. Translations from Anthropic Claude Opus 4.8, at $13.64 per million characters, have 12.2 weighted errors per 1,000 words, and those from DeepL Next Gen, at $27.50, have 38.6.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 29.8 |
| Gender neutrality | 9.7 |
| Tone of voice | 2.4 |
| Tag handling | 0.3 |
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 Galician | Class |
|---|---|---|
| Microsoft Translator | 48.5 | MT engines |
| Cohere Command A Translate | 40.9 | LLMs |
| DeepL Next Gen | 38.6 | LLM-based MT engines |
| xAI Grok 4.20 (reasoning) | 33.3 | LLMs |
| MoonshotAI Kimi K2.5 | 32.3 | LLMs |
| Translated Lara 2 | 32.3 | LLM-based MT engines |
| z.ai GLM-5 | 27.4 | LLMs |
| OpenAI GPT-4.1 | 25.7 | LLMs |
| Google Gemini 3.5 Flash | 19.1 | LLMs |
| OpenAI GPT-5.6 Sol | 13.2 | LLMs |
| Anthropic Claude Opus 4.8 | 12.2 | LLMs |
| Anthropic Claude Fable 5 | 11.5 | LLMs |
Translations from the nine LLMs on the shortlist have 11.5 to 40.9 weighted errors per 1,000 words in Galician, those from Translated Lara 2 and DeepL Next Gen, the two LLM-based MT engines, 32.3 and 38.6, and those from the MT engine Microsoft Translator 48.5. All nine shortlisted LLMs support Galician by our measured criteria. The four solutions that meet general translation quality are all on LLMs.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | 0.789 | 11 | 1.56 | 13 | No | Not shortlisted |
| OpenAI GPT-5.5 | 0.789 | 12 | 1.61 | 18 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.788 | 13 | 1.62 | 19 | No | Not shortlisted |
| Mistral Large 3 | 0.787 | 14 | 1.63 | 22 | No | Not shortlisted |
| DeepSeek V3.2 | 0.786 | 17 | 1.76 | 24 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.785 | 18 | 1.54 | 11 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.784 | 19 | 1.58 | 17 | No | Not shortlisted |
| Google Translation LLM | 0.784 | 20 | 1.79 | 26 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.784 | 21 | 1.62 | 21 | No | Not shortlisted |
| Google Gemma 4 | 0.783 | 22 | 1.62 | 20 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.783 | 23 | 1.65 | 23 | No | Not shortlisted |
| Elia Machine Translation System | 0.783 | 24 | 1.54 | 10 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.780 | 25 | 1.57 | 16 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.780 | 26 | 1.56 | 14 | No | Not shortlisted |
| MiniMax M2.5 | 0.771 | 27 | 1.98 | 29 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.766 | 28 | 1.79 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.765 | 29 | 1.83 | 27 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.756 | 30 | 2.42 | 30 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.738 | 31 | 2.62 | 31 | No | Not shortlisted |
| NiuTrans | 0.651 | 32 | 3.26 | 32 | No | Not shortlisted |
| Cohere Tiny Aya | 0.596 | 33 | 6.24 | 33 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.796 | 1 | 1.56 | 12 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | 0.795 | 2 | 1.52 | 8 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.795 | 3 | 1.50 | 5 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | 0.794 | 4 | 1.51 | 6 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.794 | 5 | 1.50 | 4 | Yes | Top 8 by both metrics |
| MoonshotAI Kimi K2.5 | 0.793 | 6 | 1.53 | 9 | Yes | Top 8 by COMET-Kiwi |
| Cohere Command A Translate | 0.793 | 7 | 1.57 | 15 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Sol | 0.792 | 8 | 1.45 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.792 | 9 | 1.51 | 7 | Yes | Top 8 by MetricX |
| Microsoft Translator | 0.790 | 10 | 1.46 | 3 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.787 | 15 | 1.86 | 28 | Yes | Real-time place |
| z.ai GLM-5 | 0.786 | 16 | 1.46 | 2 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Anthropic Claude Opus 4.8 | Adjusted LLM | 1 | 0.796 | 12 | 1.56 | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | Adjusted LLM | 2 | 0.795 | 8 | 1.52 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 3 | 0.795 | 5 | 1.50 | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 4 | 0.794 | 6 | 1.51 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 5 | 0.794 | 4 | 1.50 | Top 8 by both metrics |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 6 | 0.793 | 9 | 1.53 | Top 8 by COMET-Kiwi |
| Cohere Command A Translate | Adjusted LLM | 7 | 0.793 | 15 | 1.57 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 8 | 0.792 | 1 | 1.45 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 9 | 0.792 | 7 | 1.51 | Top 8 by MetricX |
| Microsoft Translator | Off-the-shelf NMT | 10 | 0.790 | 3 | 1.46 | Top 8 by MetricX |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 11 | 0.789 | 13 | 1.56 | Not shortlisted |
| OpenAI GPT-5.5 | Adjusted LLM | 12 | 0.789 | 18 | 1.61 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 13 | 0.788 | 19 | 1.62 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 14 | 0.787 | 22 | 1.63 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 15 | 0.787 | 28 | 1.86 | Real-time place |
| z.ai GLM-5 | Adjusted LLM | 16 | 0.786 | 2 | 1.46 | Top 8 by MetricX |
| DeepSeek V3.2 | Adjusted LLM | 17 | 0.786 | 24 | 1.76 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 18 | 0.785 | 11 | 1.54 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 19 | 0.784 | 17 | 1.58 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 20 | 0.784 | 26 | 1.79 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 21 | 0.784 | 21 | 1.62 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 22 | 0.783 | 20 | 1.62 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.783 | 23 | 1.65 | Not shortlisted |
| Elia Machine Translation System | Off-the-shelf NMT | 24 | 0.783 | 10 | 1.54 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 25 | 0.780 | 16 | 1.57 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 26 | 0.780 | 14 | 1.56 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 27 | 0.771 | 29 | 1.98 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 28 | 0.766 | 25 | 1.79 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 29 | 0.765 | 27 | 1.83 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 30 | 0.756 | 30 | 2.42 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 31 | 0.738 | 31 | 2.62 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 32 | 0.651 | 32 | 3.26 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 33 | 0.596 | 33 | 6.24 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.790 | — | 1.59 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.788 | — | 1.68 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.785 | — | 1.62 | Same model family as xAI Grok 4.20 (reasoning) |
Every model we got a translation from, as the shortlist compared it: off the shelf, without context in the request, or adjusted, with it. A model that did not make the shortlist is shown adjusted, unless off the shelf it scores higher by 0.03 or more on COMET-Kiwi. Ranks among 33 models, with each model family counted once, and the top 8 of each metric in bold. Higher is better on COMET-Kiwi, lower on MetricX. The scores rank models within this language and are not a requirement result.
COMET-Kiwi alone brought the solutions on Anthropic Claude Opus 4.8, MoonshotAI Kimi K2.5 and Cohere Command A Translate onto the list, and of these only the one on Anthropic Claude Opus 4.8 meets general translation quality. Among those MetricX alone brought, the ones on z.ai GLM-5 and Microsoft Translator, ranked 2 and 3 on MetricX, have 27.4 and 48.5 weighted errors per 1,000 words. Nearest the cut in the tail are Google Gemini 3.1 Pro Preview, at rank 11 on COMET-Kiwi, and Elia Machine Translation System, at rank 10 on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemini 3.5 Flash, one-step LLM adjusted for Standard requirements | $3.01 |
| Standard | Google Gemini 3.5 Flash, one-step LLM adjusted for Standard requirements | $3.01 |
| Advanced | OpenAI GPT-5.6 Sol, one-step with the gender-neutrality agent | $7.20 |
| Model | US dollars per million characters | Class |
|---|---|---|
| xAI Grok 4.20 (reasoning) | $1.01 | LLMs |
| MoonshotAI Kimi K2.5 | $1.13 | LLMs |
| z.ai GLM-5 | $1.7 | LLMs |
| OpenAI GPT-4.1 | $2.72 | LLMs |
| Google Gemini 3.5 Flash | $3.01 | LLMs |
| OpenAI GPT-5.6 Sol | $6.7 | LLMs |
| Microsoft Translator | $10 | MT engines |
| Anthropic Claude Opus 4.8 | $13.64 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $27.02 | LLMs |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Sonnet 5 | 0.424 | LLMs |
| Anthropic Claude Opus 4.8 | 0.423 | LLMs |
| Anthropic Claude Fable 5 | 0.422 | LLMs |
| Mistral Medium 3.1 | 0.35 | LLMs |
| Mistral Large 3 | 0.344 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.323 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.321 | LLMs |
| MoonshotAI Kimi K2.5 | 0.302 | LLMs |
| AI21 Jamba Large 1.7 | 0.299 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.292 | Open-weight LLMs |
| DeepSeek V3.2 | 0.29 | LLMs |
| z.ai GLM-5 | 0.278 | LLMs |
| Cohere Command A Translate | 0.271 | LLMs |
| Google Gemini 3.5 Flash | 0.266 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.265 | LLMs |
| Google Gemma 4 | 0.265 | Open-weight LLMs |
| OpenAI GPT-4.1 | 0.259 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.257 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.257 | LLMs |
| OpenAI GPT-5.5 | 0.256 | LLMs |
| OpenAI GPT-5.6 Sol | 0.256 | LLMs |
| OpenAI GPT-5.6 Terra | 0.256 | LLMs |
| OpenAI GPT-5.6 Luna | 0.254 | LLMs |
| Cohere Tiny Aya | 0.246 | LLMs |
| MiniMax M2.5 | 0.243 | LLMs |
With billing on tokens, the tokenizers produce a median 0.27 tokens per character of Galician, 1.1 times their count for English. The tokenizers of Anthropic Claude Fable 5 and Anthropic Claude Opus 4.8 produce 0.422 and 0.423 tokens per character, against 0.256 for OpenAI GPT-5.6 Sol and 0.266 for Google Gemini 3.5 Flash. The cheapest solution costs $3.01 per million characters under both the Basic and the Standard requirements, and $7.20 under the Advanced ones, with the gender-neutrality agent.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | 48 |
| Standard | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | 48 |
| Advanced | OpenAI GPT-5.6 Sol, one-step with the gender-neutrality agent | 35 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| Google Translation LLM | 0.5 | LLM-based MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Translated Lara 2 | 1.2 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 1.6 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.3 | MT engines |
| Elia Machine Translation System | 6.9 | MT engines |
| OpenAI GPT-5.6 Luna | 11 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 12 | LLMs |
| Cohere Tiny Aya | 13 | LLMs |
| OpenAI GPT-4.1 | 15 | LLMs |
| OpenAI GPT-5.6 Terra | 24 | LLMs |
| Anthropic Claude Haiku 4.5 | 28 | LLMs |
| Cohere Command A Translate | 33 | LLMs |
| Mistral Medium 3.1 | 33 | LLMs |
| MiniMax M2.5 | 35 | LLMs |
| Swiss AI Apertus-8B-2509 | 36 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | 41 | LLMs |
| AI21 Jamba Large 1.7 | 42 | LLMs |
| Anthropic Claude Sonnet 5 | 45 | LLMs |
| OpenAI GPT-5.5 | 46 | LLMs |
| OpenAI GPT-5.6 Sol | 48 | LLMs |
| Mistral Large 3 | 49 | LLMs |
| Anthropic Claude Fable 5 | 53 | LLMs |
| Anthropic Claude Sonnet 4.6 | 54 | LLMs |
| DeepSeek V3.2 | 55 | LLMs |
| Anthropic Claude Opus 4.8 | 56 | LLMs |
| Google Gemini 3.5 Flash | 60 | LLMs |
| NiuTrans | 60 | MT engines |
| Google Gemini 3.1 Pro Preview | 71 | LLMs |
| Alibaba Qwen3-235B-A22B | 99 | Open-weight LLMs |
| Google Gemma 4 | 135 | Open-weight LLMs |
| z.ai GLM-5 | 145 | LLMs |
| MoonshotAI Kimi K2.5 | 166 | 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 Galician.
Six of the seven real-time models in Galician are MT engines or LLM-based MT engines, and the seventh is the open-weight OpenAI gpt-oss-120b, which did not make the shortlist. The three real-time solutions on the shortlist have 32.3 to 48.5 weighted errors per 1,000 words, against 11.5 to 19.1 for the three that meet every Standard requirement. The fastest solution that meets every Advanced requirement takes 35 seconds per 1,000 words, against 48 for the fastest that meets every Standard requirement.
| 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 | two-step, with the gender-neutrality agent | 100% | ✓ | $53.46 | 102 |
| OpenAI GPT-5.6 Sol | one-step, with the gender-neutrality agent | 100% | ✓ | $7.20 | 35 |
For tone of voice in Galician we check the formal register, with vostede as the form of address. For gender neutrality we check neutral nouns and rephrasing, paired forms only when unavoidable, no typographic forms such as @ or x, and no default gender. With the gender-neutrality agent, translations from OpenAI GPT-5.6 Sol have 13.2 weighted errors per 1,000 words, as without it, and those from Anthropic Claude Fable 5 have 14.5 with two steps (translation + post-edit), against 11.5 with one step and no agent.
In Galician, the agreement of the quality-estimation pair of automatic metrics is 0.59, above its threshold of 0.53, and that of the similarity pair is 0.39, below its threshold of 0.46. Neither is a borderline case. The three LQA judges agree on a median of 96% of the sentences of the 12 shortlisted solutions. So the Intento LQA results on this page rest on LQA judges who agree, and the automatic scores on one pair of metrics out of two.
Leave feedback