Mistral Large 2.1

Mistral Large 2.1, released as mistral-large-2411 in November 2024, is the refreshed build of Mistral Large 2, Mistral AI's 123-billion-parameter flagship language model. It keeps the 128K context window and dense architecture while updating the weights and instruction tuning.

Large 2 was Mistral's answer to Llama 3.1-class models, claiming strength in code generation, mathematics, and multilingual reasoning across dozens of languages. The 2.1 revision sits between the original July 2024 Large 2 and the later Mistral Large 3, which moved to a 675B sparse MoE design.

It ships with open weights under the Mistral Research License, which permits research and non-commercial use but requires a commercial license from Mistral for production deployment. That license nuance matters: it is open-weight but not open-source in the OSI sense.