Galileo rolled out multiple product updates this week, centering on cost-efficient LLM evaluation, AI agent reliability, and security for enterprise deployments. The company introduced Luna Studio, a turnkey workflow for training small language model evaluators within customers’ own environments to cut evaluation costs that can rival or exceed inference spend.
Meet Samuel – Your Personal Investing Prophet
- Start a conversation with TipRanks’ trusted, data-backed investment intelligence
- Ask Samuel about stocks, your portfolio, or the market and get instant, personalized insights in seconds
Luna Studio is designed to generate production-ready evaluators from a few hundred labeled examples and targets latency of around 150 milliseconds per evaluation. It supports major cloud ML platforms such as Vertex AI, Azure ML, Amazon SageMaker, and private clusters, with Galileo emphasizing that it does not access customer data.
Across several posts, Galileo framed fine-tuned smaller models as cost-effective “LLM-as-judge” alternatives that can help enterprises avoid multimillion-dollar annual evaluation run rates. By treating evaluation as core infrastructure rather than an add-on, the company aims to deepen its role in the AI observability and quality assurance stack and increase recurring software revenue.
Galileo also launched Eval Engineer, an open-source skill bundle that integrates with Claude Code and Codex to support evaluation engineering for AI agents. The tool connects Galileo’s observability data with application codebases to create diagnostics, targeted fix plans, and verification steps directly in repositories, streamlining the path from issue detection to validated remediation.
A broader platform update highlighted Galileo’s Agent Control framework, which enforces granular, policy-based restrictions on tool calls to mitigate OWASP-listed threats such as tool misuse and loop amplification. In demos, the system intercepted Model Context Protocol requests in developer workflows, allowing read-only actions while blocking writes and merges before execution.
The company expanded tracing and analytics for AI and multi-agent systems, improved multimodal observability metrics, and added model pricing tools to track application and metric costs. These enhancements, coupled with expanded integrations across leading model providers, are intended to improve reliability, governance, and cost visibility for enterprises scaling AI agents.
For Galileo, the week’s developments reinforce a strategy focused on being a security, observability, and cost-governance layer for AI-heavy operations. The combination of Luna Studio, Eval Engineer, and strengthened agent controls may increase platform stickiness and relevance for regulated and cost-sensitive customers, supporting its long-term position in enterprise AI infrastructure.