Mistral AI

Mistral AI is a French AI lab. Its legal terms describe it as a limited joint-stock corporation incorporated in Paris, with registered offices at 15 rue des Halles, and its own timeline says it was founded in April 2023, hired its first employee on June 5, 2023, and released its first model, Mistral 7B, on September 27, 2023.^1,2 Mistral 7B was a 7.3-billion-parameter model released under the Apache 2.0 license.³ Its current catalog mixes licenses: its model documentation lists Mistral Small 4 and the speech model Voxtral Small under Apache 2.0, Mistral Medium 3.5 under a modified MIT license, and Mistral Large 4, announced October 6, 2026, as an open-weight multimodal model.^4,1 Its timeline dates the first Voxtral release to July 15, 2025.¹

In Cheapest LLM APIs per 1M Output Tokens (October 2026), Mistral Small 4 ranks 5th at $0.60 per 1M output tokens ($0.15 input), and Mistral Large 4 ranks 12th at its $4.18 list price ($1.36 input); on October 13, 2026, Mistral's pricing page also showed a sale price of $2.09 output for Large 4, which the ranking doesn't use.⁵ In speech, Artificial Analysis measures Voxtral Small at a 2.8% AA-WER, the lowest of the open-weights speech-to-text models it tracks, while its 24B-2507 version ranks 30th on Hugging Face's Open ASR Leaderboard at a 6.38% average; What Word Error Rate Measures, and Why Speech-to-Text WERs Disagree explains why the two leaderboards disagree.^6,7 By the ATOM Report's count of cumulative downloads, DeepSeek passed Mistral in January 2026, at 128.2 million against 124.3 million.⁸