With three-step solutions available, the best result per language has a median of ~11 weighted errors per 1,000 words, compared with ~19 in one step.
The best one-step solution comes from Anthropic Claude Fable 5 in 24 languages, OpenAI GPT-5.5 in 20 and Google Gemini 3.1 Pro Preview in 14.
With three-step solutions added, the best solution has the fewest errors in Kannada, Uzbek and Marathi, 5 to 6 weighted errors per 1,000 words, and the most in Sundanese, 24. It meets general translation quality in every language except Burmese, Korean and Sundanese.
The best solution in each language, one-step, with two-step and with three-step solutions
| Language | Best one-step solution | Weighted errors per 1,000 words | Sentences without an error | Best with two-step solutions | Weighted errors per 1,000 words | Best with three-step solutions | Weighted errors per 1,000 words | Best meets general translation quality |
|---|---|---|---|---|---|---|---|---|
| Afrikaans | OpenAI GPT-5.5 | 28.7 | 67 | OpenAI GPT-5.5, one-step | 28.7 | Anthropic Claude Fable 5, three-step | 18.8 | yes |
| Albanian | Anthropic Claude Fable 5 | 18.1 | 76 | Anthropic Claude Fable 5, one-step | 18.1 | Anthropic Claude Fable 5, one-step | 18.1 | yes |
| Amharic | OpenAI GPT-5.6 Sol | 23.1 | 74 | OpenAI GPT-5.6 Sol, one-step | 23.1 | Google Gemini 3.1 Pro Preview, three-step | 9.9 | yes |
| Arabic (MSA) | OpenAI GPT-5.5 | 16.2 | 81 | OpenAI GPT-5.5, one-step | 16.2 | OpenAI GPT-5.5, one-step | 16.2 | yes |
| Armenian | OpenAI GPT-5.6 Terra | 26.7 | 70 | OpenAI GPT-5.5, two-step | 20.1 | OpenAI GPT-5.6 Terra, three-step | 8.6 | yes |
| Assamese | OpenAI GPT-5.5 | 30.3 | 68 | OpenAI GPT-5.5, two-step | 22.7 | Google Gemini 3.1 Pro Preview, three-step | 11.5 | yes |
| Azerbaijani | OpenAI GPT-5.5 | 16.5 | 79 | OpenAI GPT-5.5, one-step | 16.5 | OpenAI GPT-5.5, one-step | 16.5 | yes |
| Basque | OpenAI GPT-5.6 Sol | 21.8 | 71 | OpenAI GPT-5.6 Sol, one-step | 21.8 | OpenAI GPT-5.6 Sol, three-step | 11.9 | yes |
| Belarusian | Anthropic Claude Fable 5 | 19.8 | 74 | Anthropic Claude Fable 5, two-step | 17.1 | OpenAI GPT-5.6 Sol, three-step | 7.6 | yes |
| Bengali | OpenAI GPT-5.5 | 17.1 | 81 | OpenAI GPT-5.5, one-step | 17.1 | OpenAI GPT-5.5, one-step | 17.1 | yes |
| Bosnian | Google Gemini 3.1 Pro Preview | 17.5 | 75 | Google Gemini 3.1 Pro Preview, one-step | 17.5 | Google Gemini 3.1 Pro Preview, one-step | 17.5 | yes |
| Bulgarian | Google Gemini 3.1 Pro Preview | 11.2 | 85 | Google Gemini 3.1 Pro Preview, one-step | 11.2 | Google Gemini 3.1 Pro Preview, one-step | 11.2 | yes |
| Burmese | OpenAI GPT-5.6 Luna | 33.3 | 69 | OpenAI GPT-5.5, two-step | 27.7 | OpenAI GPT-5.5, three-step | 13.2 | no |
| Catalan | Google Gemini 3.1 Pro Preview | 11.9 | 82 | Google Gemini 3.1 Pro Preview, one-step | 11.9 | Google Gemini 3.1 Pro Preview, one-step | 11.9 | yes |
| Cebuano | Google Gemini 3.1 Pro Preview | 34.9 | 62 | Google Gemini 3.1 Pro Preview, two-step | 34.3 | Google Gemini 3.1 Pro Preview, three-step | 11.2 | yes |
| Chinese Simplified | OpenAI GPT-5.6 Terra | 22.7 | 68 | OpenAI GPT-5.6 Terra, one-step | 22.7 | z.ai GLM-5, three-step | 13.8 | yes |
| Chinese Traditional | Anthropic Claude Opus 4.8 | 23.7 | 71 | OpenAI GPT-5.5, two-step | 21.4 | OpenAI GPT-5.5, three-step | 11.2 | yes |
| Croatian | OpenAI GPT-5.5 | 9.6 | 88 | OpenAI GPT-5.5, one-step | 9.6 | OpenAI GPT-5.5, one-step | 9.6 | yes |
| Czech | Anthropic Claude Fable 5 | 11.2 | 85 | Anthropic Claude Fable 5, one-step | 11.2 | Anthropic Claude Fable 5, one-step | 11.2 | yes |
| Danish | OpenAI GPT-5.5 | 12.2 | 84 | OpenAI GPT-5.5, one-step | 12.2 | OpenAI GPT-5.5, one-step | 12.2 | yes |
| Dutch | OpenAI GPT-5.5 | 14.5 | 81 | OpenAI GPT-5.5, one-step | 14.5 | OpenAI GPT-5.5, one-step | 14.5 | yes |
| Estonian | OpenAI GPT-5.6 Sol | 19.4 | 73 | OpenAI GPT-5.6 Sol, one-step | 19.4 | Anthropic Claude Opus 4.8, three-step | 9.2 | yes |
| Farsi | Google Gemini 3.1 Pro Preview | 12.9 | 81 | Google Gemini 3.1 Pro Preview, one-step | 12.9 | Google Gemini 3.1 Pro Preview, one-step | 12.9 | yes |
| Finnish | OpenAI GPT-5.6 Sol | 13.2 | 81 | OpenAI GPT-5.6 Sol, one-step | 13.2 | OpenAI GPT-5.6 Sol, one-step | 13.2 | yes |
| French | Anthropic Claude Fable 5 | 22.1 | 68 | Anthropic Claude Fable 5, one-step | 22.1 | Anthropic Claude Fable 5, three-step | 15.5 | yes |
| Frisian | OpenAI GPT-5.5 | 28.3 | 67 | OpenAI GPT-5.5, two-step | 24.7 | OpenAI GPT-5.5, three-step | 9.6 | yes |
| Galician | Anthropic Claude Fable 5 | 11.5 | 85 | Anthropic Claude Fable 5, one-step | 11.5 | Anthropic Claude Fable 5, one-step | 11.5 | yes |
| Georgian | Anthropic Claude Fable 5 | 16.2 | 77 | Anthropic Claude Fable 5, one-step | 16.2 | Anthropic Claude Fable 5, one-step | 16.2 | yes |
| German | Google Gemini 3.1 Pro Preview | 10.5 | 82 | Google Gemini 3.1 Pro Preview, one-step | 10.5 | Anthropic Claude Opus 4.8, three-step | 6.6 | yes |
| Greek | Anthropic Claude Fable 5 | 10.2 | 85 | Anthropic Claude Fable 5, one-step | 10.2 | Anthropic Claude Fable 5, one-step | 10.2 | yes |
| Gujarati | Anthropic Claude Fable 5 | 18.8 | 79 | Google Gemini 3.1 Pro Preview, two-step | 14.8 | Google Gemini 3.1 Pro Preview, two-step | 14.8 | yes |
| Hebrew | Anthropic Claude Fable 5 | 16.5 | 83 | Anthropic Claude Fable 5, one-step | 16.5 | Google Gemini 3.1 Pro Preview, three-step | 6.6 | yes |
| Hindi | Anthropic Claude Fable 5 | 11.5 | 84 | Anthropic Claude Fable 5, one-step | 11.5 | Anthropic Claude Fable 5, one-step | 11.5 | yes |
| Hungarian | Anthropic Claude Fable 5 | 16.2 | 81 | Google Gemini 3.1 Pro Preview, two-step | 15.2 | Google Gemini 3.1 Pro Preview, two-step | 15.2 | yes |
| Icelandic | OpenAI GPT-5.5 | 23.1 | 72 | OpenAI GPT-5.5, two-step | 15.8 | Google Gemini 3.1 Pro Preview, three-step | 7.9 | yes |
| Indonesian | Google Gemini 3.1 Pro Preview | 12.5 | 81 | Google Gemini 3.1 Pro Preview, one-step | 12.5 | Google Gemini 3.1 Pro Preview, one-step | 12.5 | yes |
| Irish | Google Gemini 3.1 Pro Preview | 25 | 66 | Google Gemini 3.1 Pro Preview, two-step | 24.7 | Google Gemini 3.1 Pro Preview, three-step | 6.9 | yes |
| Italian | OpenAI GPT-5.5 | 11.2 | 82 | OpenAI GPT-5.5, one-step | 11.2 | OpenAI GPT-5.5, one-step | 11.2 | yes |
| Japanese | Google Gemini 3.1 Pro Preview | 22.4 | 71 | Google Gemini 3.1 Pro Preview, one-step | 22.4 | Anthropic Claude Opus 4.8, three-step | 11.5 | yes |
| Kannada | OpenAI GPT-5.6 Terra | 23.7 | 72 | OpenAI GPT-5.6 Terra, two-step | 22.1 | OpenAI GPT-5.6 Terra, three-step | 4.9 | yes |
| Kazakh | OpenAI GPT-5.5 | 19.1 | 79 | OpenAI GPT-5.5, one-step | 19.1 | Google Gemini 3.1 Pro Preview, three-step | 11.5 | yes |
| Khmer | OpenAI GPT-5.6 Sol | 18.5 | 75 | OpenAI GPT-5.6 Sol, two-step | 16.8 | OpenAI GPT-5.6 Sol, two-step | 16.8 | yes |
| Korean | OpenAI GPT-5.6 Terra | 21.4 | 77 | OpenAI GPT-5.6 Terra, one-step | 21.4 | Anthropic Claude Fable 5, three-step | 13.8 | no |
| Kyrgyz | OpenAI GPT-5.5 | 20.4 | 76 | OpenAI GPT-5.5, one-step | 20.4 | Anthropic Claude Fable 5, three-step | 8.9 | yes |
| Lao | Google Gemini 3.1 Pro Preview | 16.8 | 77 | OpenAI GPT-5.6 Sol, two-step | 16.5 | OpenAI GPT-5.6 Sol, two-step | 16.5 | yes |
| Latvian | OpenAI GPT-5.5 | 16.8 | 79 | OpenAI GPT-5.5, one-step | 16.8 | Anthropic Claude Fable 5, three-step | 7.3 | yes |
| Lithuanian | Google Gemini 3.1 Pro Preview | 18.8 | 75 | Google Gemini 3.1 Pro Preview, one-step | 18.8 | DeepSeek V3.2, three-step | 8.2 | yes |
| Macedonian | OpenAI GPT-5.6 Terra | 11.9 | 82 | OpenAI GPT-5.6 Terra, one-step | 11.9 | OpenAI GPT-5.6 Terra, one-step | 11.9 | yes |
| Malay | OpenAI GPT-5.5 | 10.9 | 86 | OpenAI GPT-5.5, one-step | 10.9 | OpenAI GPT-5.5, one-step | 10.9 | yes |
| Malayalam | OpenAI GPT-5.6 Sol | 16.8 | 83 | Google Gemini 3.1 Pro Preview, two-step | 16.8 | OpenAI GPT-5.5, three-step | 9.2 | yes |
| Marathi | OpenAI GPT-5.6 Sol | 16.2 | 79 | OpenAI GPT-5.6 Sol, one-step | 16.2 | OpenAI GPT-5.6 Sol, three-step | 5.9 | yes |
| Mongolian | Google Gemini 3.1 Pro Preview | 30.7 | 67 | OpenAI GPT-5.5, two-step | 25.4 | OpenAI GPT-5.5, three-step | 11.2 | yes |
| Nepali | OpenAI GPT-5.5 | 19.8 | 75 | OpenAI GPT-5.5, one-step | 19.8 | Google Gemini 3.1 Pro Preview, three-step | 7.3 | yes |
| Norwegian | OpenAI GPT-5.5 | 9.9 | 86 | OpenAI GPT-5.5, one-step | 9.9 | OpenAI GPT-5.5, one-step | 9.9 | yes |
| Odia (Oriya) | OpenAI GPT-5.6 Luna | 23.1 | 73 | OpenAI GPT-5.6 Luna, one-step | 23.1 | Google Gemini 3.1 Pro Preview, three-step | 12.9 | yes |
| Pashto | OpenAI GPT-5.6 Sol | 26.4 | 68 | OpenAI GPT-5.6 Sol, one-step | 26.4 | OpenAI GPT-5.5, three-step | 10.9 | yes |
| Polish | OpenAI GPT-5.6 Terra | 13.5 | 80 | OpenAI GPT-5.6 Terra, one-step | 13.5 | Anthropic Claude Opus 4.8, three-step | 8.6 | yes |
| Portuguese (European) | Anthropic Claude Fable 5 | 14.8 | 69 | Anthropic Claude Fable 5, one-step | 14.8 | Anthropic Claude Sonnet 4.6, three-step | 7.9 | yes |
| Punjabi | OpenAI GPT-5.6 Sol | 16.8 | 79 | OpenAI GPT-5.6 Sol, one-step | 16.8 | OpenAI GPT-5.5, three-step | 13.5 | yes |
| Romanian | Anthropic Claude Fable 5 | 16.2 | 82 | Anthropic Claude Fable 5, one-step | 16.2 | Anthropic Claude Fable 5, one-step | 16.2 | yes |
| Russian | Anthropic Claude Fable 5 | 12.9 | 85 | Anthropic Claude Fable 5, one-step | 12.9 | Anthropic Claude Fable 5, one-step | 12.9 | yes |
| Serbian | Anthropic Claude Fable 5 | 17.8 | 76 | Anthropic Claude Fable 5, one-step | 17.8 | Anthropic Claude Fable 5, one-step | 17.8 | yes |
| Sinhala | Anthropic Claude Fable 5 | 17.1 | 78 | Anthropic Claude Fable 5, one-step | 17.1 | OpenAI GPT-5.6 Luna, three-step | 11.2 | yes |
| Slovak | OpenAI GPT-5.6 Terra | 11.9 | 84 | OpenAI GPT-5.6 Terra, one-step | 11.9 | OpenAI GPT-5.6 Terra, one-step | 11.9 | yes |
| Slovenian | Anthropic Claude Fable 5 | 10.9 | 85 | Anthropic Claude Fable 5, one-step | 10.9 | Anthropic Claude Fable 5, one-step | 10.9 | yes |
| Spanish (European) | OpenAI GPT-5.6 Terra | 9.9 | 86 | OpenAI GPT-5.6 Terra, one-step | 9.9 | OpenAI GPT-5.6 Terra, one-step | 9.9 | yes |
| Sundanese | OpenAI GPT-5.5 | 35.3 | 62 | OpenAI GPT-5.5, one-step | 35.3 | Anthropic Claude Sonnet 4.6, three-step | 24.1 | no |
| Swahili | Anthropic Claude Fable 5 | 36.3 | 66 | Anthropic Claude Fable 5, one-step | 36.3 | Anthropic Claude Fable 5, three-step | 9.6 | yes |
| Swedish | OpenAI GPT-5.5 | 12.5 | 82 | OpenAI GPT-5.5, one-step | 12.5 | OpenAI GPT-5.5, one-step | 12.5 | yes |
| Tagalog | OpenAI GPT-5.6 Terra | 21.4 | 77 | OpenAI GPT-5.6 Terra, one-step | 21.4 | OpenAI GPT-5.5, three-step | 7.6 | yes |
| Tajik | OpenAI GPT-5.6 Sol | 22.7 | 70 | OpenAI GPT-5.6 Sol, two-step | 21.1 | OpenAI GPT-5.6 Sol, three-step | 8.2 | yes |
| Tamil | OpenAI GPT-5.6 Sol | 10.5 | 84 | OpenAI GPT-5.6 Sol, one-step | 10.5 | OpenAI GPT-5.6 Sol, one-step | 10.5 | yes |
| Tatar | Google Gemini 3.1 Pro Preview | 30 | 68 | Google Gemini 3.1 Pro Preview, two-step | 27.7 | Google Gemini 3.1 Pro Preview, three-step | 8.6 | yes |
| Telugu | Anthropic Claude Fable 5 | 14.5 | 81 | Anthropic Claude Fable 5, one-step | 14.5 | Google Gemini 3.1 Pro Preview, three-step | 8.6 | yes |
| Thai | Anthropic Claude Opus 4.8 | 19.8 | 73 | Anthropic Claude Opus 4.8, one-step | 19.8 | OpenAI GPT-5.6 Sol, three-step | 9.2 | yes |
| Turkish | Anthropic Claude Fable 5 | 12.2 | 86 | Anthropic Claude Fable 5, one-step | 12.2 | Anthropic Claude Fable 5, one-step | 12.2 | yes |
| Ukrainian | Anthropic Claude Fable 5 | 9.9 | 84 | Anthropic Claude Fable 5, one-step | 9.9 | Anthropic Claude Fable 5, one-step | 9.9 | yes |
| Urdu | Anthropic Claude Fable 5 | 18.5 | 78 | Anthropic Claude Fable 5, one-step | 18.5 | Anthropic Claude Fable 5, three-step | 7.3 | yes |
| Uzbek | OpenAI GPT-5.5 | 22.4 | 73 | Google Gemini 3.1 Pro Preview, two-step | 17.5 | OpenAI GPT-5.5, three-step | 5.6 | yes |
| Vietnamese | Anthropic Claude Fable 5 | 13.8 | 82 | Anthropic Claude Fable 5, one-step | 13.8 | Anthropic Claude Fable 5, one-step | 13.8 | yes |
| Welsh | Google Gemini 3.1 Pro Preview | 12.9 | 83 | Google Gemini 3.1 Pro Preview, one-step | 12.9 | Google Gemini 3.1 Pro Preview, one-step | 12.9 | yes |
| Yiddish | Anthropic Claude Fable 5 | 44.2 | 65 | Anthropic Claude Fable 5, two-step | 34.6 | Anthropic Claude Fable 5, three-step | 8.9 | yes |
Units — Sentences without an error: %
The best solution has the fewest weighted errors per 1,000 words (minor 1, major 5, critical 10) among the solutions with an Intento LQA result: first the one-step solutions, then with the two-step solutions (translation + post-edit) and the three-step solutions (translation + LQA + repair) added. General translation quality is met with no critical error, at most three major errors in the four texts and at most 20 weighted errors per 1,000 words.