Details
- Mistral AI announced Mistral Large 4, a general-purpose model designed to run agents that collect information and produce completed deliverables across complex workflows.
- On AutomationBench, it completed or scored highest across 657 simulated business workflows involving Gmail, Google Sheets, Slack, and Salesforce.
- Mistral said Large 4 ranked ahead of Kimi K3, MiMo-V2.6-Pro, and DeepSeek V4 Pro on the workplace-app benchmark.
- The model is positioned for operational tasks that require coordinating multiple applications rather than generating isolated responses.
- On Harvey’s Legal Agent Benchmark, Mistral said Large 4 was the top open-source model, indicating progress on legal research and document workflows.
- The announcement highlights competition in agentic AI, where performance depends on reliable tool use, multi-step execution, and completed outcomes.
- Mistral’s official announcement describes Large 4 as an open-weight, general-purpose multimodal model with a mixture-of-experts architecture and a 1-million-token context window.
Impact
Large 4 puts Mistral into a stronger position in enterprise-agent competition by emphasizing measurable performance across workplace software and legal workflows. Its reported AutomationBench lead over several major open-model rivals could encourage businesses to evaluate open-weight systems for multi-application automation, while the legal result strengthens Mistral’s case in regulated professional services. The claims remain vendor-reported and should be assessed against independent evaluations as the model becomes broadly available.
