Mistral Large 3 Delivers 92% of GPT-5.2 Performance at 15% of the Cost

Mistral Large 3, released under Apache 2.0 license on January 25, delivers 92% of GPT-5.2's performance at approximately 15% of the cost, according to independent benchmarks published February 1.

Performance Analysis

In comprehensive testing across reasoning, code generation, and knowledge tasks, Mistral Large 3 scored within 92% of GPT-5.2 while costing $0.80 per million tokens versus GPT-5.2's $5.00—an 84% cost reduction.

Efficiency Architecture

The model uses a sparse mixture-of-experts architecture with 675B total parameters but only 56B active per token. This efficiency enables near-GPT performance while maintaining reasonable inference costs.

Open Source Advantage

Unlike GPT-5.2, Mistral Large 3's Apache 2.0 license allows commercial use, modification, and self-hosting. Organizations can run the model on their infrastructure, keeping data private while avoiding per-token API costs.

Market Implications

The performance-to-cost ratio challenges the viability of closed-source models for many use cases. If open models can deliver 92% of proprietary performance at 15% of the cost, the economics shift dramatically toward open alternatives.