Researchers report higher agent task completion with reusable AI components

Researchers report in an arXiv preprint that their HEELO framework increased Terminal-Bench 4.0 task resolution from 37.3% under Codex to 55.5% using the same model backbone. The framework assembles task-specific networks of reusable components, each equipped with a resident language model that applies a resource in context. HEELO checks results against evidence from the environment and revises components implicated in failures. On 46 tasks from the researchers' own platform, Claude Opus 5.5 with Claude Code and GPT-6.1 Sol with Codex each completed 9 tasks alone, rising to 39 and 35 respectively when connected to HEELO. Those platform results are part of the authors' evaluation.