Factory Launches Droid 1.0 With Script Mode to Cut Its AI Agents' Token Use
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
Factory introduces Droid 1.0, an agent for programming in enterprise settings, with a Script mode in which the model writes a short program that bundles hundreds of tool calls into one request. In its tests with Claude Opus 5.5, Droid cuts tokens by 17% and time per passed task by 26% compared with Claude Code. Droid, with GPT-6.1 Sol, saves 12% of tokens against Codex, but needs 10% more time.
Factory has launched Droid 1.0, the new version of its coding agent for enterprises, with a Script mode and figures that, according to the company itself, cut token spending, tokens being the units used to measure how much an AI model is used.
Droid is what Factory calls its "harness", that is, everything around the model: the instructions, the tools and the loop that runs them. In Script mode, the model writes a small JavaScript program that runs in an isolated environment and makes hundreds of tool calls in a single request. That way the instructions and the history are sent once, instead of being resent every few calls.
Factory says the first request of a session went from 20,887 to 13,055 tokens, but the company itself notes that this compares an earlier version of Droid running Claude Opus 4.8 with the current one on Claude Opus 5.5, and that the tokenizers, the way tokens are counted, differ. Measured in characters, the drop is 35%. Most of the saving comes from the tool definitions sent up front, which went from 11 to 2; Factory does not attribute it to Script mode.
In its own tests on Terminal-Bench 4, a benchmark of terminal tasks, Droid with Claude Opus 5.5 uses 17% fewer tokens than Claude Code and takes 26% less time per passed task. With GPT-6.1 Sol it uses 12% fewer tokens than Codex, though it takes 10% longer. These are the company's own results.
Droid 1.0 has been available since October 29 as a desktop app for macOS and Windows, in the terminal and with an SDK for building it into other applications.
Why it matters · analysis and opinion
For anyone paying for coding agents, cost is set in large part by how many times instructions and history are resent, and Factory goes after exactly that: fewer requests and fewer tokens per task. But the comparisons are internal, made on a single benchmark, and against Codex the token saving comes with more time, so cost and speed do not always improve together. The data is published by Factory itself alongside the launch of its product, and the post includes no independent verification. A company that is interested can try Droid on its own tasks and compare tokens, time and result before switching tools.
Official source: Factory · Written with the help of AI: how we make the news
