The State of Translation Automation 2026 Performance

For most models, translation into the slowest language takes over three times as long as into the median language.

With an MT engine, the slowest languages differ from those of the LLMs. With Yandex Cloud Translate v2, the slowest five are Ukrainian, Kyrgyz, Azerbaijani, Tajik and Tatar, and translation into any language takes at most twice the median time.

With Translated Lara 2, the slowest languages are those of the LLMs, at a fraction of their time. Translation into Sinhala, Lao and Odia takes 7.7 to 9.4 seconds per 1,000 words with Translated Lara 2, against 93 to 144 seconds per 1,000 words with xAI Grok 4.20 (reasoning).

Translation time in the median and slowest languages, by model

Model	Type	Median language, s per 1,000 words	Slowest language, s per 1,000 words	Slowest language	Slowest against median	Languages
Translated Lara 2	LLM-based MT engine	2	30.1	Tamil	15.1×	97
OpenAI gpt-oss-120b	open-weight LLM	2.73	9.1	Lao	3.3×	100
Yandex Cloud Translate v2	MT engine	4.11	8.1	Tatar	2.0×	83
OpenAI GPT-5.6 Luna	LLM	15.97	49	Lao	3.1×	100
xAI Grok 4.20 (non-reasoning)	LLM	20.35	5,029.7	Hausa	247.1×	100
OpenAI GPT-4.1	LLM	28.09	137.4	Sinhala	4.9×	100
OpenAI GPT-5.6 Terra	LLM	39.49	123.1	Odia (Oriya)	3.1×	100
Anthropic Claude Haiku 4.5	LLM	45.33	109.6	Khmer	2.4×	100
Swiss AI Apertus-8B-2509	open-weight LLM	50.82	258.1	Igbo	5.1×	100
Mistral Medium 3.1	LLM	53.48	1,756.1	Amharic	32.8×	100
MiniMax M2.5	LLM	58.39	708.7	Lao	12.1×	100
Cohere Command A Translate	LLM	63.59	167.2	Amharic	2.6×	100
AI21 Jamba Large 1.7	LLM	64.03	149.7	Sinhala	2.3×	100
xAI Grok 4.20 (reasoning)	LLM	65.02	231.3	Yoruba	3.6×	100
DeepSeek V3.2	LLM	65.27	143.5	Welsh	2.2×	100
Google Gemini 3.5 Flash	LLM	67.63	101.4	Odia (Oriya)	1.5×	100
Mistral Large 3	LLM	78.68	528.9	Odia (Oriya)	6.7×	100
Anthropic Claude Sonnet 5	LLM	78.77	266.3	Khmer	3.4×	100
Anthropic Claude Fable 5	LLM	87.56	335.9	Lao	3.8×	100
Anthropic Claude Sonnet 4.6	LLM	89.72	178.5	Odia (Oriya)	2.0×	100
OpenAI GPT-5.5	LLM	90.39	360.8	Odia (Oriya)	4.0×	100
Anthropic Claude Opus 4.8	LLM	96.27	367.7	Malayalam	3.8×	100
Google Gemini 3.1 Pro Preview	LLM	99.24	420.3	Nepali	4.2×	100
OpenAI GPT-5.6 Sol	LLM	99.38	419.9	Burmese	4.2×	100
Google Gemma 4	open-weight LLM	224.41	449.1	Tatar	2.0×	100
z.ai GLM-5	LLM	237.07	5,929.6	Odia (Oriya)	25.0×	100
MoonshotAI Kimi K2.5	LLM	323.15	658.6	Lao	2.0×	100

Translated without context in the request, each text timed separately.

Languages among the ten slowest and the ten fastest of each model

Language	Time against the median language of the model	Models with it among the ten slowest	Models with it among the ten fastest	Models
Odia (Oriya)	2.7	23	0	26
Lao	2.43	22	0	27
Sinhala	2.33	17	0	27
Amharic	2.28	19	0	27
Khmer	1.86	16	0	27
Uyghur	1.84	16	0	26
Yiddish	1.75	13	1	27
Burmese	1.69	14	0	27
Malayalam	1.64	11	0	27
Kannada	1.52	6	0	27
Gujarati	1.5	8	0	27
Armenian	1.49	7	0	27
Punjabi	1.49	10	0	27
Telugu	1.44	4	0	27
Tamil	1.4	4	0	27
Pashto	1.39	1	1	26
Tatar	1.39	7	0	27
Yoruba	1.36	6	0	26
Georgian	1.35	7	1	27
Kyrgyz	1.32	2	0	27
Assamese	1.3	1	1	26
Irish	1.24	4	1	27
Scottish Gaelic	1.22	0	1	27
Kinyarwanda	1.21	1	0	26
Tajik	1.21	2	0	27
Turkmen	1.2	3	0	26
Mongolian	1.19	1	0	27
Zulu	1.19	2	0	27
Sesotho	1.18	3	1	26
Somali	1.18	1	0	26
Wolof	1.17	4	1	26
Malagasy	1.16	2	0	27
Maltese	1.16	2	1	27
Shona	1.16	1	0	26
Xhosa	1.16	1	0	27
Igbo	1.15	1	0	26
Greek	1.14	1	0	27
Kazakh	1.14	1	0	26
Nepali	1.13	2	0	27
Belarusian	1.12	2	0	27
Kurdish (Kurmanji)	1.12	3	0	25
Hausa	1.11	2	0	26
Marathi	1.11	1	0	27
Welsh	1.09	3	0	27
Icelandic	1.02	1	0	27
Thai	1.02	2	2	27
Uzbek	1.02	0	0	27
Basque	1.01	0	0	27
Bengali	1.01	1	1	27
Urdu	1.01	0	1	27
Albanian	1	0	1	27
Azerbaijani	1	1	0	27
Frisian	0.99	0	0	25
Hebrew	0.98	1	1	27
Latvian	0.98	0	0	27
Macedonian	0.96	0	0	27
Cebuano	0.94	0	1	27
Estonian	0.94	1	0	27
Swahili	0.94	0	1	27
Sundanese	0.93	1	0	27
Finnish	0.92	0	0	27
Lithuanian	0.92	0	0	27
Haitian Creole	0.91	0	0	27
Hindi	0.91	2	2	27
Hungarian	0.91	1	0	27
Bulgarian	0.88	0	1	27
Farsi	0.86	0	4	27
Slovenian	0.86	0	0	27
Tagalog	0.85	0	1	27
Slovak	0.83	0	2	27
Javanese	0.82	0	4	27
Arabic (MSA)	0.81	0	2	27
Turkish	0.81	0	2	27
Croatian	0.8	0	1	27
Ukrainian	0.8	1	3	27
Afrikaans	0.79	0	1	27
Polish	0.78	0	2	27
Romanian	0.78	0	1	27
Serbian	0.78	1	5	27
Bosnian	0.77	0	1	27
Danish	0.77	0	2	27
Czech	0.74	0	2	27
Japanese	0.74	0	6	27
Swedish	0.74	0	6	27
Vietnamese	0.73	0	11	27
Catalan	0.72	0	3	27
Korean	0.71	0	8	27
Norwegian	0.71	0	4	26
Malay	0.69	0	6	27
Dutch	0.67	0	10	27
Galician	0.67	0	9	27
Portuguese (European)	0.65	0	17	27
Russian	0.64	0	14	27
Spanish (European)	0.64	0	19	27
Indonesian	0.62	0	18	27
Italian	0.62	0	17	27
Chinese Simplified	0.61	0	21	27
French	0.6	0	15	27
German	0.6	0	14	27
Chinese Traditional	0.59	0	19	26

Units — Time against the median language of the model: × the model median

The time per 1,000 words in a language divided by the median over the languages of the same model, the median across models. Models timed by text in 80 or more languages, translated without context in the request.

For most models, translation into the slowest language takes over three times as long as into the median language.

Translation time in the median and slowest languages, by model

Languages among the ten slowest and the ten fastest of each model

Automate translation to meet your requirements