The AWS Strands Agents team has released Strands Harness, an open-source, general-purpose agent harness available under the Apache 2.0 license for Python and TypeScript. The tool is designed to bridge the gap between prototyping agents in environments like Claude Code or Codex and deploying them with custom loops.
- The harness reports a 28% lower token cost compared to rival harnesses across six benchmarks, including ALFWorld, GAIA, and Terminal-Bench 2.1, while maintaining comparable accuracy.
- It features built-in context management rules that truncate tool results over 1,500 tokens, trigger summarization at 85% context usage, and handle recovery within the loop.
- Benchmarks using Claude Fable 5 on Terminal-Bench 2.1 showed Strands Harness costing $56.29 with 69.7 accuracy, significantly cheaper than Claude Code ($248.05, 61.8 accuracy) while outperforming it in score.
- The system supports running on current reasoning models via Amazon Bedrock, Anthropic, OpenAI, Google, Ollama, or LiteLLM, and includes shell, file, and web tools out of the box.
The release aims to provide developers with a deployable, general-purpose agent framework that reduces operational costs through efficient context handling without sacrificing performance.