OpenAI has shipped a multi-agent v2 update that adds support for the Luna model, a release aimed at making AI workflows more efficient and cheaper to run across a range of applications. The update enables seamless task delegation between agents, reducing the need for human coordination in multi-step processes.
Task delegation takes center stage
The core of the v2 update is how agents hand off work to each other. In earlier versions of OpenAI's multi-agent system, agents typically waited for a central controller to assign tasks. The v2 update changes that by letting agents pass tasks directly to one another. A job that starts with one agent can now flow through a chain of specialized agents without a person stepping in at each stage.
That shift is designed to cut down on the manual oversight that often bogs down AI-driven workflows. Instead of a human mapping out every step, agents can figure out which of their peers should handle the next piece. The company says this makes workflows more efficient, though it hasn't detailed the technical mechanics of how the handoffs are orchestrated.
Luna joins the agent lineup
The update also brings Luna into the fold. Luna is a model that can now be used within multi-agent setups, joining the existing roster of models available through OpenAI's platform. Developers get another option when designing agent workflows, but the company hasn't said what Luna is specifically optimized for. Its inclusion suggests OpenAI sees a role for it in the kind of distributed, task-oriented work that multi-agent systems handle.
For teams already building on OpenAI's API, this means they can slot Luna into their agent pipelines without reworking the whole system. The v2 update is backward-compatible, though the company didn't specify whether that's the case—just that the update is out now.
Efficiency gains with cost in mind
The efficiency gains aren't just about speed. OpenAI says the update enables cost-effective operations across diverse applications. By letting agents work together more fluidly, companies can potentially reduce the number of API calls needed to finish a task, or trim the time spent on manual oversight. That could matter for teams running AI at scale, where even small efficiency improvements translate into real savings.
The "diverse applications" part is worth attention. Multi-agent systems are used in customer support, data analysis, software development, and other fields. The v2 update is designed to work across those different contexts, though OpenAI doesn't single out which use cases it expects to benefit most.
What's still unclear
The update is live, but open questions remain. How well Luna performs in real-world multi-agent workloads is untested outside OpenAI's own environment. The company hasn't released benchmarks or case studies yet, so early adopters will be the ones to validate the efficiency claims.
It's also unclear whether the new delegation model will require changes to how developers structure their agents. Some teams may need to rework their existing pipelines to take full advantage of the handoff feature. For now, anyone using OpenAI's platform can start experimenting with the v2 update and Luna. Whether it delivers on its promises will become clearer as more deployments go live.
