Factory has released Droid 1.0, featuring programmatic tool calling by default to improve efficiency and reduce costs for software development tasks. The update introduces a Script mode where the model writes JavaScript to execute multiple tool calls within a secure sandbox, keeping intermediate logs and raw files out of the main context window. This approach allows a single model request to drive hundreds of tool calls concurrently while significantly lowering token usage and latency.
The coding agent harness supports various models, including Claude Opus 5.5 and GPT-6.1 Sol, and operates across macOS, Windows, a terminal UI, and a Node SDK. In Factory's internal evaluations on Terminal-Bench 4, Droid 1.0 achieved a 65.2% pass rate using Claude Opus 5.5, outperforming several competing harnesses while reducing costs by up to 24 percent. Additional structural improvements include a leaner base prompt with a 37 percent reduction in input tokens, optimized prompt caching, and bounded autonomy safeguards.
Why it matters: Factory Droid 1.0 demonstrates how programmatic tool execution can scale AI agent capabilities while lowering operational costs and API overhead.