Mozilla AI has made Mistral Large 4 available through Otari, letting teams already using the gateway redirect workloads with a model-selector change while keeping their existing policy, tracing and cost controls.
Model and claimed performance
Mistral Large 4, which Mistral calls “Le Chonk,” is a natively multimodal mixture-of-experts model with 1.05 trillion total parameters. Mozilla AI says it activates 49 billion parameters per token, an architecture intended to offer trillion-parameter capacity at inference costs closer to those of a mid-size dense model.
Mistral’s published evaluations, as summarized by Mozilla AI, position Large 4 ahead of other U.S. and European open-weight models on aggregate benchmarks. Mistral also reports leading open-model results for enterprise workloads in cybersecurity, finance and manufacturing, plus strong coding and agentic performance. Those are vendor-reported results, so teams should test the model against their own workload requirements.
Multimodal workloads
Mozilla AI highlights complex documents, engineering drawings and satellite imagery as potential inputs for Large 4. The stated use case is an agentic workflow in which a model inspects visual material, reasons over it and takes an action through connected tools.
Routing through Otari
Otari sits between applications or agents and the models and tools they call. Mozilla AI says centralizing routing, policy and observability there means an existing Otari deployment can add Large 4 without adopting a separate SDK, credential set or governance configuration.
Models use a provider: model selector, making a switch to Large 4 a string-level routing change. Mozilla AI also says OpenAI SDK users can retain their code by pointing the SDK’s base URL to Otari. Provider keys are managed in the gateway, which can run the model alongside other Mistral models and other providers, with routing and provider failover handled at that layer.
Policy and measurement
Otari applies gateway-level policies when traffic is routed to Large 4, according to Mozilla AI. Its agent hooks can run deterministic checks based on recorded evidence, such as whether a protected file was edited or required tests passed. It also supports LLM-as-a-judge checks for qualitative questions such as whether work follows an architecture or meets acceptance criteria.
Model calls, agent actions and policy decisions are recorded in a common history with traces, timestamps and per-request costs. That gives teams a way to compare Large 4 with a current model on latency, spending and policy outcomes using the same operational data. Otari is open source and provider-agnostic, and Mozilla AI says it can be self-hosted or used as a managed platform. Source: Mozilla AI
Definition. Otari is a provider-agnostic gateway between applications or agents and the models and tools they call.
Key takeaways
- Mistral Large 4 is available through Mozilla AI's Otari gateway.
- Existing Otari users can select Large 4 through provider:model routing.
- Gateway-level policies, traces and per-request cost tracking remain available for routed traffic.
- Otari can manage provider keys, routing and provider failover at the gateway layer.
- Teams can use common operational data to compare Large 4 with a current model on latency, spending and policy outcomes.
FAQ
How can existing Otari users route workloads to Mistral Large 4?
Mozilla AI says models use a provider:model selector, so switching to Large 4 is a string-level routing change.
Can OpenAI SDK users use Otari with Mistral Large 4?
Mozilla AI says OpenAI SDK users can retain their code by pointing the SDK base URL to Otari.
What controls remain available when routing traffic to Large 4?
Otari applies gateway-level policies and records calls, actions and policy decisions with traces, timestamps and per-request costs.
What types of inputs does Mozilla AI highlight for Large 4?
Mozilla AI highlights complex documents, engineering drawings and satellite imagery as potential inputs.
