Google's Madlad 400 3B Mt is a multilingual machine translation model based on the T5 architecture, trained on 1 trillion tokens covering over 450 languages. It is particularly well-suited for machine translation and multilingual NLP tasks, supporting over 400 languages.
Input
Output
Context
512
Max Output
-
Parameters
2.9B
Input Modalities
Output Modalities
Some values come from web research and may be incorrect. Report incorrect data
| Precision | Weights | One request at 512 tokens | Smallest GPU that fits |
|---|---|---|---|
| FP16 / BF16 | 5.48 GB | 5.80 GB | 1x NVIDIA T4 (16 GB) |
| INT8 | 2.74 GB | 2.92 GB | 1x NVIDIA T4 (16 GB) |
| INT4 | 1.37 GB | 1.49 GB | 1x NVIDIA T4 (16 GB) |
Computed by Inferbase from the 2,940,374,016 parameters in the creator's Hugging Face repository; one request adds its KV cache, activations and framework overhead, with the layer layout estimated from the parameter count.
Specification sources
As of Nov 10, 2026.
Answered from the facts on this page, with their sources.
Madlad 400 3B Mt has a context window of 512 tokens (source: Creator or provider API). A provider may serve less; Inferbase routes on what each provider serves.
Yes. Madlad 400 3B Mt is an open-weight model under the Apache 2.0 license. You can download and self-host it within the license terms.
Madlad 400 3B Mt has 2,940,374,016 parameters (2.9B), as stated in the creator's Hugging Face repository.
Strengths: Open weights under a permissive license (apache-2.0); Supports 417 languages. Limitations: Short context window: 512 tokens. Each is computed from the model's stated facts and benchmark results.
Madlad 400 3B Mt accepts Text and produces Text. The Capabilities card lists the API features it supports, such as function calling and structured output.
No benchmark source has published a result for Madlad 400 3B Mt yet, so it is shown as unmeasured, not estimated.
Not on Inferbase Managed today. Add your own provider key or endpoint that serves Madlad 400 3B Mt, and Inferbase runs it on your account with the same routing and logs.
The weights take 5.5 GB at FP16 / BF16, 2.7 GB at INT8, 1.4 GB at INT4, computed from the 2,940,374,016 parameters its creator publishes. Each request adds its KV cache and the framework's overhead. The hardware section shows one request and the smallest GPU that fits; the GPU planner sizes it for your context length and traffic.
The model card of Madlad 400 3B Mt lists 417 languages: English, Russian, Spanish, French, German, Italian, Portuguese, Polish, Dutch, Vietnamese, Turkish, Swedish, Indonesian, Romanian, Czech, Chinese, Hungarian, Japanese, Thai, Finnish, Persian, Ukrainian, Danish, Greek, Norwegian, Bulgarian, Slovak, Korean, Arabic, Lithuanian, Catalan, Slovenian, Hebrew, Estonian, Latvian, Hindi, Albanian, Malay, Azerbaijani, Serbian, Tamil, Croatian, Kazakh, Icelandic, Malayalam, Marathi, Telugu, Afrikaans, Galician, Filipino, Belarusian, Macedonian, Basque, Bangla, Georgian, Mongolian, Bosnian, Uzbek, Urdu, Swahili, Cantonese, Nepali, Kannada, Kara-Kalpak, Gujarati, Sinhala, Welsh, Esperanto, Latin, Armenian, Kyrgyz, Tajik, Irish, Maltese, Burmese, Khmer, Tatar, Somali, Kurdish, Pashto, Punjabi, Kinyarwanda, Lao, Hausa, Divehi, Western Frisian, Luxembourgish, Central Kurdish, Malagasy, Scottish Gaelic, Amharic, Uyghur, Haitian Creole, Ancient Greek, Hmong, Sindhi, Javanese, Māori, Turkmen, Cebuano, Yiddish, Bashkir, Faroese, Odia, Xhosa, Sundanese, Kalaallisut, Nyanja, Samoan, Shona, Corsican, Zulu, Igbo, Yoruba, Papiamento, Southern Sotho, Hawaiian, Assamese, Occitan, Chuvash, Mizo, Tetum, Swiss German, Yakut, Breton, Romansh, Sanskrit, Tibetan, Oromo, Northern Sami, Chechen, cnh, Iloko, Hiligaynon, Udmurt, Ossetic, Ganda, Tigrinya, Venetian, Tsonga, Tuvinian, Kabardian, Ewe, Iban, Avaric, Khasi, Tongan, Tswana, Northern Sotho, Fijian, Zaza, Akan, Adangme, otq, Dzongkha, Buriat, cfm, Lingala, Mari, Guarani, Karachay-Balkar, Walloon, Fiji Hindi, yua, Sranan Tongo, Waray, Romany, Bikol, Pampanga, Sango, Luba-Katanga, Adyghe, kbp, Syriac, Latgalian, Erzya, iso, Kachin, Bhojpuri, Aymara, Kumyk, Quechua, Zhuang, Pangasinan, ngu, Venda, pck, Zapotec, tyz, hui, Batak Toba, tzo, Tiv, ksd, Konkani, Minangkabau, Old English, nhe, bgp, Nzima, nnb, Navajo, No linguistic content, bci, Komi, Newari, mps, Southern Altai, meu, Betawi, Fon, Inuktitut, abt, Makhuwa-Meetto, mnw, Tuvalu, dov, Klingon, Hiri Motu, Cornish, Western Mari, meo, Crimean Tatar, mbt, emp, Acehnese, ium, mam, gym, Maithili, Seselwa Creole French, Pohnpeian, ubu, fip, Kʼicheʼ, Manx, Kuanyama, btx, ape, Chuukese, rcf, Shan, tzh, Moksha, ppk, Swati, Gagauz, cab, Krio, Sena, Ibibio, tbz, bru, enq, Acoli, cuk, Kimbundu, Wolof, Qʼeqchiʼ, qub, tab, bts, Kosraean, rwo, cak, tuc, Bulu, cjk, Gilbertese, Saterland Frisian, tsg, quh, Makasar, Mapuche, Balinese, jiv, sja, Yapese, Tulu, toj, twu, Kalmyk, amu, rmc, hus, Nias, kjh, Bambara, guh, Masai, acf, Central Dusun, ksw, bzj, Dinka, zne, Madurese, msi, Magahi, mkn, Kongo, lhu, Chamorro, qvi, Marshallese, djk, Susu, Morisyen, srm, Dyula, ctu, gui, Palauan, inb, Bislama, Manipuri, Wayuu, Jamaican Creole English, Wolaytta, jac, Basaa, Gorontalo, skr, nyu, noa, sda, gub, Nogai, cni, Teso, tdx, sxn, rki, South Ndebele, Arpitan, alz, taj, Northern Luri, cce, Rundi, jvn, hvn, nij, dwr, izz, msm, bus, ktu, Cherokee, maz, tzj, suz, knj, bim, gvl, bqc, tca, Pijin, prk, laj, mel, qxr, niq, ahk, shp, hne, spp, Komi-Permyak, Kinaray-a, quf, luz, agr, tsc, mqy, gof, gbm, miq, Zarma, Awadhi, bjj, qvz, sjp, tll, Rajasthani, kjg, bgz, quy, cbk, akb, Ojibwa, ify, mey, Kashmiri, cac, Bodo, qup, syl, jax, Fula, ber, tks, trp, mrw, adh, smt, Serer, ffm, qvc, mtr, Obolo, Afar, noe, nut, gyn, kwi, xmm and msb. That is what Google states; the catalog does not test language quality.
Start on the free plan with 5,000 requests a month, no card required.