Optimize your team's price-performance with hosted open weight models

This post introduces three new hosted open weight models (Kimi K3, GLM 5.3, and MiniMax M3) to the GitLab Duo Agent Platform, expanding model choice for optimizing cost and performance. It details the cost-performance advantages of open weight models, providing specific call-per-credit multipliers for each new model and comparing them to existing default models. The post also outlines administrative controls for model selection, deployment options (GitLab-managed, self-hosted, hybrid), and the security and risk management processes for vendor-hosted models, emphasizing a zero data retention policy with Fireworks AI. It highlights how these models can be used for different agentic tasks based on their strengths (large context, long-horizon coding, cost efficiency at scale).