French artificial intelligence startup Mistral has unveiled a new model named Mistral Large 4, also referred to as Le Chonk, ending a five-month release hiatus, UNN reported, citing Reuters. The model is scheduled to become generally accessible on October 27.
Speaking at a presentation in Abu Dhabi, Mistral Chief Executive Officer Arthur Mensch stated that the model outperforms Chinese counterparts in select evaluation metrics, particularly in cybersecurity, arguing against assertions that European developers cannot compete. Mensch did not specify which Chinese models or benchmarks were used in the comparison.
Mistral reported that Large 4 narrows the performance gap with frontier systems across software development, finance, geospatial analysis, manufacturing, and product engineering. The startup trained the model on its own technical infrastructure. Before the public rollout, designated cybersecurity specialists and government authorities will receive access to an evaluation version with fewer safety guardrails to assess capabilities.
Pierre Stock, Mistral's vice president of science, told Reuters that the system ranks among the most capable open-weight models, enabling clients to download parameters and run workloads on private infrastructure. While leading US models are typically closed-source, numerous Chinese systems use open-weight distribution. Stock also highlighted collaboration across the supply chain with corporate backers such as ASML and Samsung, noting the model can design computer chips. In August, Mistral also began hosting third-party offerings, including the GLM 5.3 model from Beijing-based Z.ai.
Sources
- Українські Національні Новини (УНН) · 2026-10-06

