Unreal Labs has introduced Unreal Agent, an asynchronous-first harness designed to improve the cost-efficiency and responsiveness of AI agents by optimizing how they manage tool calls.
Product LaunchesUnreal LabsUnreal Agent
Unreal Labs Unveils Unreal Agent to Optimize AI Agent Cost-Efficiency via Asynchronous Tool Calls
The harness addresses the overhead caused by agents spending excessive time and tokens on managing tool-related tasks such as waiting, polling, and heartbeats. By managing tool calls asynchronously, Unreal Agent allows the underlying model to focus on reasoning while tool execution occurs in the background. This approach enables users to steer the agent without waiting for tool calls to finish and allows the agent to schedule additional work between model calls.
In benchmarks using GPT-6 Astra xhigh, Unreal Agent demonstrated significant efficiency gains, achieving up to 40% cost savings compared to Codex and up to 20% compared to Pi in real workloads. The company noted that the harness design achieves these results by minimizing the model's cognitive load and using a simplified, token-optimized approach to tool outputs.
Sources
- Unreal Agent （Hacker News Frontpage, 2026-09-22）
- Unreal Agent