이미지: METAL LAB 생성
Summary
- Nous Research unveiled a new "browser use mode" for its agent Hermes
- It replaced 12 browser tools with a single tool powered by browser_use CLI 3.0
- The company said internal tests showed token usage dropped 48-66% with no loss in accuracy
- 발표 주체
- Nous Research
- 기존 도구 수
- 브라우저 도구 12개
- 변경 후
- browser_use CLI 3.0 기반 단일 도구
- 토큰 절감폭
- 48~66% (자체 테스트 기준)
- 정확도 변화
- 저하 없음
Twelve tools became one
Nous Research announced it has applied a new "browser use mode" to its agent Hermes. Previously, the agent relied on twelve separate tools for browser actions like clicking, scrolling, and typing. These have now been replaced with a single tool powered by browser_use's CLI 3.0. The company explained that "instead of calling a tool for every click, the agent now writes a script." According to its internal tests, token usage dropped by 48-66% with no decrease in accuracy.
Why tool calls were so expensive
For an AI agent to operate a web browser, it traditionally has to send the LLM a separate request for every single action — "click this button," "type this text into this field," and so on. This approach is called a "tool call," and each request requires resending the full documentation (schema) for all twelve available tools to the model. Since this entire set of instructions gets attached even for a single click, the wasted tokens — the basic unit an LLM uses to process text, and the basis for pricing and speed — pile up quickly as conversations grow longer.
Nous Research took a different approach: having the agent write a code script that bundles multiple actions together and executes them at once. This eliminates the need to resend the twelve-tool documentation for every click. This "represent actions as code" approach echoes a direction several researchers in the agent-design community have recently been exploring. Nous Research is known as the developer behind the open-source LLM family Hermes, and this announcement also appears to be an experiment aimed at the open-source ecosystem. browser_use is a browser automation framework that lets AI agents click, type, and navigate real websites, and its CLI (command-line interface) 3.0 version now serves as the engine powering Hermes's new mode.
So what actually changes
Browser-operating AI agents are an area where both cloud providers and startups have been racing to launch products recently. The persistent problem has been cost — multi-step web page operations consume large amounts of tokens, making them burdensome to deploy in production. If this improvement holds up, it means the same task could be handled with less than half the tokens, which could help lower the operating costs of services that run agents continuously. However, this is based on Nous Research's own internal testing, and it has not been confirmed whether the results have been verified by any independent benchmark.
