AWS open-sources Strands Harness for building general-purpose AI agents · Digg

AWS open-sources Strands Harness for building general-purpose AI agents

AWS says its defaults cut benchmark cost versus Claude Code and Codex, but the company’s own tests also found DeepSeek Harness cheaper on matched runs.

1 Source, 6h ago, first seen 6h ago

TLDR

AWS has released Strands Harness, an open-source Python and TypeScript foundation for general-purpose AI agents. It bundles file, shell and web tools with context management, memory, persistent sessions, prompt caching and delegation, while allowing developers to swap model providers and customize the defaults. AWS says its own six-benchmark comparison found Strands Harness 45% cheaper than Claude Code and Codex at broadly comparable accuracy. The New Stack notes that the advantage falls to 28% when DeepSeek Harness is included, and that the figures are vendor-run benchmarks rather than independent validation.

Featured Source

Amazon Web Services has open-sourced Strands Harness, a configurable foundation for building general-purpose AI agents that can run locally or be deployed to the cloud, The New Stack reports.

The project sits above AWS’s existing Strands Agents SDK. It supplies a working agent with file, shell and web tools plus built-in context management, memory, persistent sessions, prompt caching and delegation to other agents. Developers can install Python or TypeScript packages and export configurations from the Strands command-line interface.

The model provider can be changed

Bedrock is the default model route, but AWS says developers can switch to Anthropic, OpenAI, Google or a local Ollama model. Tools, prompts, context behavior and session storage are also replaceable.

That makes the release less a new foundation model than a packaged set of decisions around the agent loop. AWS’s commercial path is optional: teams can run the open-source harness independently, or use the company’s AgentCore service to host and operate it.

AWS claims lower benchmark costs

AWS averaged results across six benchmarks and compared score with cost per task. The company says Strands Harness was 45% cheaper than Claude Code and Codex with broadly comparable accuracy.

The comparison changes when more rivals are included. DeepSeek Harness was about 14% cheaper than Strands on matched runs, reducing Strands’ claimed advantage across the wider set to 28%, according to The New Stack. On Terminal-Bench 2.1, AWS says Strands running Fable 5 cost $56.29 across 89 trials and scored 69.7, while DeepSeek Harness cost $40.30 and scored 59.5.

Those are AWS-run benchmarks, not independent validation. They do, however, illustrate the company’s argument that context truncation, compaction and recovery behavior can materially change an agent’s cost even when the underlying model is held constant.

AWS open-sources Strands Harness for building general-purpose AI agents

AWS says its defaults cut benchmark cost versus Claude Code and Codex, but the company’s own tests also found DeepSeek Harness cheaper on matched runs.

1 Source, 6h ago, first seen 6h ago

TLDR

Featured Source

Amazon Web Services has open-sourced Strands Harness, a configurable foundation for building general-purpose AI agents that can run locally or be deployed to the cloud, The New Stack reports.

The model provider can be changed

Bedrock is the default model route, but AWS says developers can switch to Anthropic, OpenAI, Google or a local Ollama model. Tools, prompts, context behavior and session storage are also replaceable.

AWS claims lower benchmark costs

AWS averaged results across six benchmarks and compared score with cost per task. The company says Strands Harness was 45% cheaper than Claude Code and Codex with broadly comparable accuracy.

1 Source

@thenewstackAWS open-sources an AI agent it says is 45% cheaper than Claude Code and Codex https://thenewstack.io/aws-strands-harness-agent/?taid=6ab4e69eabb846000197999f&utm_campaign=trueanthem&utm_medium=social&utm_source=twitter

Combined views

601

1 Source, first seen 6h ago

3 likes

AWS open-sources Strands Harness for building general-purpose AI agents

TLDR

The model provider can be changed

AWS claims lower benchmark costs

AWS open-sources Strands Harness for building general-purpose AI agents

TLDR

The model provider can be changed

AWS claims lower benchmark costs

1 Source

Combined views

The model provider can be changed

AWS claims lower benchmark costs

The model provider can be changed

AWS claims lower benchmark costs

Combined views

Sentiment

Summary

Sentiment

Summary

Sentiment

Summary

1 Source