Agent-Native Primitives improve real-world task resolution for coding agents
Original title: The Last Mile to the Real World: Agent-Native Primitives as Reusable Building Blocks for Solving Real-World Tasks
HEELO framework using eight primitive families raises Terminal-Bench 4.0 resolution from 37.3% to 55.5% and reduces task time on heelo.dev
Agent-Native Primitives equip resources with resident LLMs for end-to-end real-world task handling. The HEELO framework composes these primitives into task-specific graphs and revises them on failure. Evaluations show gains on Terminal-Bench, ToolBench, and GAIA-Text-103 across more than twenty benchmarks.
- Dev-Primitives raise Terminal-Bench 4.0 to 55.5%
- Harness Primitives reach 72.2% on Terminal-Bench 2.0
- Tool Primitives improve ToolBench pass rate to 75.1%
- Navigation Primitives boost GAIA-Text-103 to 58.3%
From the original article
1 min read at huggingface.co
How can an agent handle a real-world task end to end? Frontier models with their own harnesses still fall short on real-world benchmarks: on Terminal-Bench 4. 0, GPT-5.
By Haibo Jin, Peng Kuang, Xiaopeng Yuan · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.