# OpenAI ships multi-agent v2 update adding support for Luna model > Source: > Published: 2026-08-16 19:36:33+00:00 Matthew Kupfer/The Standard # OpenAI ships multi-agent v2 update adding support for Luna model The update lets GPT-5.6 Sol and Terra delegate tasks to Luna, fixing a compatibility gap that had frustrated developers for weeks. OpenAI has rolled out a major upgrade to its multi-agent framework that finally lets its top-tier AI models boss around its fastest one. Multi-Agent v2 now enables GPT-5.6 Sol and Terra to delegate tasks directly to Luna, the speed-optimized variant that was previously locked out of the party due to a version mismatch. The fix addresses what had become a growing pain point for developers building complex AI workflows. Luna was stuck on Multi-Agent v1 classification, which meant Sol and Terra, both v2 models, couldn’t hand it tasks. ## What changed and why it matters OpenAI’s GPT-5.6 lineup is structured as a three-tier system, each variant optimized for different use cases. Sol handles heavy-lift reasoning tasks, the kind of work that requires deep analysis and multi-step logic chains. Terra sits in the middle, balancing capability against cost. Luna is the speedster, built for high-volume applications where response time and affordability matter more than raw cognitive horsepower. The problem was that these models couldn’t actually work together the way developers expected. A Sol-based primary agent trying to delegate a quick lookup or data-formatting task to Luna would hit a wall. The v1/v2 incompatibility meant task delegation simply failed under certain conditions, forcing developers to either route everything through a single model tier or build clunky workarounds. Multi-Agent v2 removes that barrier. A primary agent running on Sol or Terra can now spin up Luna as a subagent, assign it a task, and receive the output, all within a single orchestrated workflow. The practical implications are significant. A developer building a customer service platform, for example, could use Sol for complex complaint resolution requiring nuanced reasoning, while routing simple FAQ responses to Luna at a fraction of the cost and latency. Before this update, that kind of tiered architecture within a single agent system wasn’t reliably possible. ## The compatibility headache that preceded this The version mismatch between Luna and the other GPT-5.6 variants wasn’t just an inconvenience. It was actively disrupting project workflows for developers who had built their systems expecting full interoperability across the model family. Community feedback from developer forums suggests the frustration was widespread enough to make this fix a priority. OpenAI reportedly conducted thorough reliability testing before shipping the update, a nod to the fact that multi-agent systems are only as good as their weakest coordination layer. That said, the rollout isn’t entirely smooth. Developers have flagged that tools like Codex still need additional configuration adjustments to fully leverage v2 multi-agent functionality. So while the core delegation pipeline now works, the broader tooling ecosystem hasn’t completely caught up yet. **Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our [Editorial Policy](https://cryptobriefing.com/editorial-policy/).