AI News Feed
Market watch
Xi to Make First U.S. State Visit Since Obama Era; Trade Truce and AI on AgendaWhatsApp Expands Apple Watch Emoji Reactions to the Full watchOS Emoji SetAWS Strands Agents Team Releases Open-Source Strands Harness, Claims 28% Lower Token Cost at Comparable AccuracyiOS 27.2 beta 2 code points to continued work on Apple's unannounced iPhone anti-snatching featureGlobal AI Glasses Shipments Jump 263% in First Half of 2026 as Meta Tightens GripKRICT adds graphitic carbon nitride to lithium-ion cathodes, lifting power delivery in early tests
Xi to Make First U.S. State Visit Since Obama Era; Trade Truce and AI on AgendaWhatsApp Expands Apple Watch Emoji Reactions to the Full watchOS Emoji SetAWS Strands Agents Team Releases Open-Source Strands Harness, Claims 28% Lower Token Cost at Comparable AccuracyiOS 27.2 beta 2 code points to continued work on Apple's unannounced iPhone anti-snatching featureGlobal AI Glasses Shipments Jump 263% in First Half of 2026 as Meta Tightens GripKRICT adds graphitic carbon nitride to lithium-ion cathodes, lifting power delivery in early tests
Products & Applications

AWS Strands Agents Team Releases Open-Source Strands Harness, Claims 28% Lower Token Cost at Comparable Accuracy

The Strands Agents team at AWS released Strands Harness, an Apache 2.0 general-purpose agent harness for Python and TypeScript. The team reports 28% lower token cost than rival harnesses across six benchmarks at comparable accuracy.

AI News Feed · 2026-09-22 07:39

The team describes a harness as the system around a model: the loop, tools, context handling, memory, and recovery. Strands already exposed those building blocks through the Strands Harness SDK; Strands Harness packages them into working defaults. The result is a general-purpose agent, not a coding agent. Out of the box, create_harness() returns an agent that runs on a current reasoning model through Amazon Bedrock, Anthropic, OpenAI, Google, Ollama, or LiteLLM.

The default agent ships with shell, file read/write/edit, and web tools rather than a bespoke tool for each task. It offloads bulky tool results to files and caches reused parts of each request. It keeps long-term memory across runs and can resume a conversation from a session ID. It delegates open-ended subtasks to a built-in helper agent and tracks multi-step work with a checklist. It also loads Agent Skills when it finds them. A bundled skills file can help a coding agent generate deployment configuration for AWS, GCP, Azure, Cloudflare, and Modal.

The Strands Agents team ran distributed benchmarking on Amazon EC2 with Harbor, the evaluation framework from the Terminal-Bench creators. The score is an average across six benchmarks: ALFWorld, ContextBench, GAIA, WebShop, τ²-bench, and Terminal-Bench 2.1. Cost is average dollars per task. Rivals listed by the team are Claude Code, Codex, oh-my-pi, OpenCode, and DeepSeek Harness. The team reports 28% lower cost than other harnesses running the same Claude or GPT models at near-equal accuracy. DeepSeek Harness was the most token-efficient harness overall, running about 14% cheaper than Strands Harness, but it scored lower on every benchmark; the chart footnote states that including it brought the overall savings figure down to 28%. The highest-scoring point on the chart is Claude Opus 5 on Strands Harness, near 85%.

The clearest head-to-head uses Claude Fable 5 on Terminal-Bench 2.1, with 89 trials per harness. Strands Harness cost $56.29 and scored 69.7. Oh-my-pi cost $86.83 and also scored 69.7. OpenCode cost $73.42 and scored 66.3. Claude Code cost $248.05 and scored 61.8. DeepSeek Harness cost $40.30 and scored 59.5. Against Claude Code, Strands Harness cost 77% less and scored 7.9 points higher. Oh-my-pi matched its 69.7 accuracy at 54% higher cost. DeepSeek Harness was cheaper still, but trailed by 10.2 points. The team also noted that two other open-source harnesses performed well on cost and accuracy against Claude Code.

Strands Harness ships defaults for prompt caching and context management. The team says context management largely drove both token efficiency and accuracy. Three rules do the work: tool results over about 1,500 tokens get truncated; summarization, or compaction, triggers when context usage passes 85%; and context recovery runs inside the loop if the window overflows. This matches recent independent research. The HarnessTax study compared Claude Code, Codex CLI, and Pi across seven models. It found that harness choice barely moved success rates, while the same model reached similar success at up to five times the cost. The Strands researchers say a follow-up paper on their benchmarks is coming.

Developers can install the harness with pip install strands-harness or npm install @strands-agents/harness. They can pick a model by name or point the harness at a local Ollama model. The Strands CLI, installed with npm install @strands-agents/strands-cli, lets users prototype an agent in plain English. In the team's demo, the agent was asked to add the Playwright MCP server and measure video load latency on a blog post. Running /export then produced the harness code, with the Playwright MCP included, as a Python or TypeScript zip. The CLI itself is built on Strands Harness. Strands engineer Gautam Sirdeshmukh also used it to build a desktop app that starts Strands Harness runs remotely. Customization goes deep: users can override any default, swap models, add tools, or replace components down to the Strands Harness SDK. Because the harness is a library dependency, the agent prototyped on a laptop is the same one embedded in production.