Droid 1.0 Writes Tool Calls as Scripts: 65.2% Pass Rate, $4.73 Per Successful Task

Factory has officially launched Droid 1.0. As reported by Asif Razzaq from MarkTechPost, this is a model-agnostic coding Agent harness that uses programmatic tool calls by default, with its underlying logic running on Script Mode.

A harness can be interpreted as an execution wrapper sitting outside the model: it connects to tools, manages context, runs tests, and handles evaluation. The model-agnostic design means this wrapper is not bound to any specific model, so users do not need to overhaul the entire workflow when switching to a different foundation model. Droid 1.0 sets programmatic tool calls as the default, and all tool invocation tasks are organized and executed via scripts. This direction is highly practical for coding Agents, as terminal-based tasks often require consecutive calls, combined commands, and targeted output processing.

Factory released test results from its self-hosted Terminal-Bench 4: when paired with Claude Opus 5.5, Droid reached a 65.2% pass rate, with a cost of $4.73 per successfully completed task. Terminal-Bench 4 assesses users' ability to finish multi-step operations in a terminal environment, so performance metrics only deliver meaningful insights when viewed together with corresponding cost data. It is worth noting that this is a vendor-run self-test, sourced from Factory's own test pipeline. While the pass rate figure is impressive, the pricing accounting rules are more notable. The $4.73 figure only covers a single passed task: costs from failed retries, total API call volume, and expenses incurred by failed tasks are not specified in this public release. Always check the full cost breakdown alongside pass rate data to avoid misleading skewed comparisons.

Original Article: Meet Factory Droid 1.0: A Model-Independent Coding Agent Harness Built on Script Mode