Glm

Mistral released the weights for Mistral Large 4, codenamed Le Chonk, claiming it rivals top open models like GLM-5.3 on the Artificial Analysis Coding Agent Index. The release includes benchmark scores the company says are competitive with leading open-weight models.

Model Release • Techmeme

Epoch AI researchers say the open-weight model GLM 5.3 turned public security fixes into working Firefox exploit chains with zero human input in under 48 hours, compromising six Firefox releases before updates reached users. The work cost about $1,000 in API fees per exploit and did not trigger safeguards from any provider tested.

Security • The Decoder