Mistral Large 4 updated weights and lower pricing on AI Gateway

Mistral Large 4 on Vercel’s AI Gateway has received updated weights that enhance agentic coding and defensive cybersecurity while improving token efficiency on coding tasks. The update also strengthens spreadsheet creation and editing, web research capabilities, and resistance to prompt injection. Users already accessing the model via AI Gateway do not need to switch model identifiers, as the existing string now points to the updated version. The changes take effect automatically without additional configuration.

To use Mistral Large 4 in coding agents like fx, Pi, Claude Code, Codex, or Cursor, users must install the latest Vercel CLI and run setup, then select the model in their agent. The process is streamlined to minimize disruption, and the model remains accessible through the AI Gateway’s unified API. Vercel provides a coding agents guide to assist with integration steps.

AI Gateway serves as a unified interface for model access, offering built-in features such as usage tracking, cost management, and configurable retries and failover mechanisms. It supports performance optimizations to achieve higher uptime than provider baselines. The service includes Zero Data Retention support, custom reporting, and budget controls for API keys, alongside routing rules for flexible model selection.

AI Gateway mirrors provider pricing without markup and does not impose platform fees on inference costs, including for Bring Your Own Key requests. Users can review the full catalog of language models available on the platform. The updated Mistral Large 4 pricing applies to input, cached input, and output tokens, reflecting the changes in the provider’s cost structure.