Factory has released Droid 1.0, the biggest update yet to its model-independent coding agent harness, with programmatic tool calling now on by default. On Factory’s own Terminal-Bench 4 runs, Droid scores 65.2% with Claude Opus 5.5 at $4.73 per passed task — beating Cursor, Devin, Claude Code and OpenCode at lower cost.
Factory Droid 1.0 (Credits: MarkTechPost)
Factory Droid 1.0: Programmatic Tool Calling by Default
Factory defines a harness as everything around the model — the prompts, the tools, how tool results return, and the loop. Users pick a model or let Factory Router choose a cost-efficient mix, and it works with any supported model including your own API keys.
The key change in Droid 1.0: instead of every tool output landing in the model’s context, the model writes a JavaScript program that runs tools inside a sandbox (with no file, network or system access of its own). Raw files, search results and logs stay inside the program, and only printed output returns to the model. The result: a unit test fix that took 10 model calls now takes 5, and a Linear cleanup of 151 issues ran in just 2 requests versus a modeled 36. On Claude Opus 5.5, Script mode cuts Droid’s cost by 15% versus direct tool calling, and it is on by default for 10 models including GPT-6.1 Sol and Kimi K3.
How Droid 1.0 Compares to Cursor, Devin and Claude Code
Factory ran each harness against Droid on the same model, with three trials per task from October 23 to 28, 2026. On Claude Opus 5.5, Droid scored 65.2% versus Cursor and OpenCode at 62.1%, Devin at 59.6% and Claude Code at 57.6% — while costing up to 24% less. On GPT-6.1 Sol, Droid hit 57.6% at $0.79 per passed task against Codex at 54.0%, OpenCode at 51.0% and Devin at 45.0%. Speed-wise, Droid takes 16% to 26% less time per passed task on Opus 5.5, and 17% fewer tokens than Claude Code.
Other improvements: the base prompt shrank 37% to 13,055 tokens, only 2 tool definitions ship up front (down from 11), compaction summaries doubled to 16,000 tokens, and app first paint dropped from 1,839 ms to 488 ms. Most changes came from Factory’s internal Agent Factory loop, where agents read failures and propose changes that engineers review and merge. Pricing starts from $20/month.
Also Read: Vercel Labs Open-Sources Agent-Desktop: AI Agents Can Drive Your Desktop Apps Without Screenshots
Source: MarkTechPost — read the full story there.
