Prime Agent achieves 13x faster input and 83% less memory usage

GLM-5.3: GLM-5.3 is Z.ai's latest flagship text-only Mixture-of-Experts language model optimized for complex coding, long-horizon agentic workflows, and vulnerability discovery. It supports extended context windows and selectable reasoning effort levels, with all improvements derived from post-training on the GLM-5.2 base. The model supplied the tokens processed by the Prime Agent swarm during the self-rewrite operation described in the news. Prime Agent: Prime Agent is an open-source self-improving coding and research harness developed around Recursive Language Model (RLM) and Continual Harness abstractions. It enables persistent Python REPL environments where models can treat context as variables, delegate to sub-agents programmatically, and modify their own prompts, skills, memory, and sub-agent specifications during runtime. In this news, Prime Agent orchestrated a multi-agent swarm to rewrite its own codebase in Rust, delivering measurable gains in speed and resource efficiency. Prime Intellect: Prime Intellect is a company focused on distributed AI infrastructure and open superintelligence development. It created and open-sourced Prime Agent as a general-purpose harness for long-horizon autonomous coding and research tasks. The entity is directly responsible for the Rust rewrite project and the multi-agent infrastructure tested in the reported experiment. Performance: Rewriting Prime Agent in Rust produced substantial improvements in input processing speed and memory efficiency relative to the prior TypeScript version. Infrastructure: Prime Intellect maintains scalable sandbox and messaging infrastructure that supports large-scale multi-agent coordination across persistent sessions. Agent Capabilities: Prime Agent supports self-orchestration of agent swarms, allowing the system to manage its own evolution through coordinated sub-agent interactions and durable state management.