A trillion-parameter model with open weights and a 1-million-token context: Mistral Large 4 is reportedly the top open model on Cognition’s FrontierCode.
Mistral AI released the model, nicknamed “Le Chonk,” with 1.05 trillion total parameters and 52 billion active parameters. It uses a mixture-of-experts (MoE) design, which activates only part of the model for computation.
That distinction matters for deployment. The active parameter count describes the portion used in computation; it does not establish the memory, hardware or operating cost needed to run the model.
Before evaluating its open weights, businesses need a short checklist:
- Hardware: confirm memory needs, accelerator configuration and operating requirements.
- Licensing: check which uses and modifications the license permits.
- Workload testing: measure speed, capacity and task quality on the intended setup.
The reported FrontierCode ranking is a signal to validate on the workloads the business intends to run.
The practical next step is to pair the model’s hardware requirements and license terms with measurements from the tasks it would actually handle.