Mistral Large 4
Points and comments are a snapshot, not live.
Mistral AI releases Large 4, an open-weight multimodal model with 49B active parameters.
Mistral Large 4 is a state-of-the-art, open-weight, general-purpose multimodal model using a granular Mixture-of-Experts architecture. It has 49B active parameters, 1.05T total parameters, and a 1.6B vision encoder. It features a 1M token context window. Pricing: $0.68/M input tokens, $0.07/M cached input tokens, $2.09/M output tokens. It supports structured outputs, function calling, document Q&A, chat completions, batching, and agents.
Mistral Large 4 is in public preview as of October 6, 2026.
What commenters are saying
Commenters are cautiously positive, noting Mistral is the only non-Chinese open-weights frontier model, which matters for European sovereignty and cybersecurity. Some benchmarks are said to rival GLM 5.3 and show strong vision and cyber performance. Skeptics question whether Mistral advances the state of the art, pointing to its position behind top-tier models. Others argue its value is strategic, not purely technical. There is speculation about identity with Space Bunny Alpha, and some praise the 50% launch discount.
Several commenters emphasize the importance of non-American, non-Chinese AI options for Europe, citing strategic dependencies and trust concerns.