Mozilla AI has added Mistral Large 4 to Otari, its open-source gateway for routing and governing model traffic. Existing Otari users can select the new model without adopting another SDK or rebuilding provider-specific policy and observability layers.
Mistral calls the model Le Chonk. It uses a mixture-of-experts architecture with 1.05 trillion total parameters but activates 49 billion for each token. The model accepts multiple types of input and is aimed at work involving complex documents, engineering drawings, satellite imagery, coding and agents.
Mistral reports leading aggregate results among US and European open-weight models and strong performance in cybersecurity, finance and manufacturing. Mozilla correctly frames those figures as vendor benchmarks that customers should test against their own traffic, latency and cost requirements.
Otari centralizes provider keys, traces, cost records and policy checks. Deterministic checks can enforce explicit rules, while model-based judges assess work requiring interpretation. The gateway can be self-hosted or used as a managed service. Availability through a common layer reduces integration work, but it does not validate Mistral’s quality claims or remove the need to test agent behavior after a model switch.