Mistral models
All 30 Mistral models on OpenDunes, priced in Algerian Dinars.
30 models
Mistral's flagship speech-to-text model that turns audio into transcripts with word timestamps and speaker labels, with context biasing for domain-specific vocabulary
The Voxtral Mini derivative that only transcribes, taking audio from meetings, voice notes and podcasts and returning the text through a standard transcription request
Mistral's compact speech-to-text model that turns audio into transcripts with word timestamps and speaker labels, and can be steered with context biasing to catch domain terms
Mistral AI's text embedding model from late 2023, whose 1024-dimension vectors capture how passages relate, for semantic search and RAG; Codestral Embed is the version for code
Turns written text into natural-sounding speech through a library of preset voices, taking emotion and pace tags so narration and spoken content carry the right tone and rhythm
Mistral's 3B Voxtral Mini for speech and audio understanding, handling transcription, translation and compact audio processing; Voxtral Small does the same work with 24B parameters
The transcription route to Voxtral Small 24B, for transcribing, translating and understanding audio where the larger model's capacity helps, as opposed to the 3B Voxtral Mini
Codestral Embed turns code, or the plain-language questions that look for it, into a fingerprint of its meaning, so search finds matching code. Use it to index a code repository so a coding assistant can find the right files and functions. Each input can be up to about 8 pages, and the result is numbers rather than text, so pair it with a chat model to answer.
Voxtral Mini TTS turns text into natural-sounding speech and returns audio. It can clone a voice without extra training, and it comes with 30 preset voices from four speakers in English, British English and French, each in moods such as cheerful, sad or angry. Input is limited to about 4 pages of text per request.
A 12B model made by Mistral with NVIDIA, released under Apache 2.0, that chats and calls functions in 11 languages such as French, Arabic and Hindi over a 128K context
Mistral Large 4 takes text and pictures and answers in text, reading up to about a thousand pages and writing up to about 262 pages per reply. Its focus is reasoning, coding and automated work, with tool use, exact-form answers and a thinking-effort setting you can raise or lower per request. Pick it for whole-codebase or long-document jobs.
Mistral Small 3.2 24B Instruct is the update to Small 3.1 and takes text and pictures, answering in text. Compared with 3.1 it follows instructions more closely, repeats itself less, stops endless replies and uses tools and fills in exact forms more reliably. It reads about 256 pages.
Mistral Small 4 folds three earlier Mistral strengths into one model: step-by-step thinking from Magistral, picture reading from Pixtral and coding helpers from Devstral. You send text and pictures, up to about 262 pages, and get text back, with tool use, exact-form answers and a setting for how hard it thinks.
Mistral Small 3 (2501) is a text model for fast, everyday chat and drafting, released under an open Apache 2.0 licence so you can also run it yourself. It reads about 33 pages at once, answers in text only and cannot use tools here. Mistral reports 81% accuracy on a widely used knowledge test.
Mistral Medium 3.1, the updated Medium 3 for enterprises that want frontier-level coding, STEM reasoning and multimodal work near Claude Sonnet 3.7 and Llama 4 Maverick, deployable on-prem or privately
Devstral 2 is a very large coding model for automated work across a whole code project: it explores files, changes several at once while keeping the overall design in mind, notices when a run fails and retries with a fix. Typical jobs are bug fixing and updating legacy systems. It reads about 262 pages, can use tools and fill in an exact form you give it, and comes under a modified MIT licence.
The smallest Ministral 3, a tiny 3B model with vision for light text and image work that still calls tools and returns structured output; the 8B and 14B are its larger siblings
The largest Ministral 3, a 14B model with vision whose results come close to the larger Mistral Small 3.2 24B, calling tools and returning structured output; the 8B and 3B are its smaller siblings
The middle Ministral 3, a balanced 8B model with vision for text and image work that calls tools and returns structured output, bigger than the 3B and smaller than the 14B
Mistral Medium 3.5 is a large model taking text, pictures and files and answering in text. It is the Medium line's pick for automated work: it uses several tools in a row reliably, stays on long tasks, and lets you set thinking effort per request. Its picture reading copes with any size or shape of image, and you can download it to run yourself.
Mistral Large 3 is a very large model you can download under an open Apache 2.0 licence and also run yourself. It takes text, pictures and files and answers in text, reading about 262 pages at once, and it can use tools and fill in exact forms you give it. Pick it for demanding general work where an open licence matters.
Codestral 2508 is Mistral's code model for fast, repeated edits inside an editor: you send the code before and after a gap and it fills the middle, corrects errors and writes tests. It reads about 256 pages of text or files and answers in text. Pick it for autocomplete where waiting counts.
Mistral Medium 3 is a mid-size model for coding, science and maths reasoning and company-specific adaptation, taking text, pictures and files and answering in text. It reads about 131 pages at once, and it can be deployed in a hybrid setup, on your own servers or inside a private cloud. It is the first release of the Medium 3 line.
Mistral Large is the Large 2 release, version 2407, which you can download under a Mistral licence. You send text or files and get text back, covering reasoning, code, filling in exact forms and chat in dozens of languages including Arabic, Hindi and Chinese, and more than 80 programming languages. The Mistral Large 2407 entry is the same version under its dated name.
Mistral Small 3.1, a 24B model that adds image understanding to Small 3, with a 128K context, function calls and dozens of languages, suited to local, privacy-sensitive deployments; 3.2 followed it
Voxtral Small, Mistral Small 3 with audio input added and its text performance kept, so one model with tool calls can transcribe, translate and understand speech as well as read and write text
Mistral Large 2407 is the dated name of Mistral Large 2, for work that must stay on exactly this version. You send text or files and get text back, covering reasoning, code, filling in exact forms and chat in dozens of languages including Arabic, Hindi and Chinese, and more than 80 programming languages.
Mixtral 8x22B Instruct is Mistral's own instruction-following version of the Mixtral 8x22B model. It takes text and files, up to just over 65 pages, and answers in text, and it can use tools and fill in exact forms. Mistral lists maths, coding and reasoning as strengths, in five European languages including French and German.
Mistral's 24B regional model for the Middle East and South Asia, trained on curated regional data to answer in Arabic and in Indian languages such as Tamil and Malayalam, with tool calls and file input
Mistral Medium 3.5 is a large model taking text, pictures and files and answering in text. It is the Medium line's pick for automated work: it uses several tools in a row reliably, stays on long tasks, and lets you set thinking effort per request. Its picture reading copes with any size or shape of image, and you can download it to run yourself.