Pins matching tag:Mistral

Mistral AI releases Mistral Large 2, a 123B model with a 128k-token context that it says matches GPT-4o and Claude 3 Opus on code and beats Llama 3.1 405B on code and math with under a third of the parameters.

Mistral AI releases Ministral 8B and Ministral 3B, 'les Ministraux', edge models with up to 128k context; the 8B scores 65.0 on MMLU (pretrained) and 70.9 on Arena Hard, priced at $0.10 and $0.04 per million tokens.

Mistral AI releases Mistral Small 3, a latency-optimized 24B model under Apache 2.0 that it says matches Llama 3.3 70B instruct at more than three times the speed, with over 81% on MMLU.

Mistral AI releases Mistral Small 3.1, a 24B Apache 2.0 model with image understanding and a 128k-token context window that Mistral says beats Gemma 3 and GPT-4o Mini at 150 tokens per second.

Mistral AI releases Mistral Small 3.2, a 24B Apache 2.0 update to Small 3.1 that lifts Arena Hard v2 from 19.56% to 43.1% and cuts infinite generations from 2.11% to 1.29%.

Mistral AI releases Mistral 3: Mistral Large 3, a 675B-parameter sparse mixture-of-experts model with 41B active parameters under Apache 2.0, and three dense Ministral 3 models (3B, 8B and 14B).

Mistral AI releases Mistral Medium 3.5, its first merged flagship: a 128B dense open-weight model with a 256K context window that scores 77.6% on SWE-Bench Verified and costs $1.50 per million input tokens and $7.50 per million output tokens.

Mistral AI releases Mistral Large 4 in public preview, a natively multimodal mixture-of-experts model of about 1 trillion parameters trained on its own cluster, with open weights promised by the end of October 2026.