Hermes cut 12 browser tools down to 1, token usage dropped 66%. Nous Research announced a tool update for Hermes Agent on X today. The focus was a complete overhaul of browser operations (Browser Use). Hermes originally had 12 browser tools—navigation, clicking, typing—

💡 What You Will Learn

Hermes cut 12 browser tools down to 1, token usage dropped 66%. Nous Research announced a tool update for Hermes Agent on X today. The focus was a complete overhaul of browser operations (Browser Use)

Hermes Cuts 12 Browser Tools Down to 1, Token Usage Drops 66%

Nous Research announced a tool update for Hermes Agent on X today. The focus this time is a complete overhaul of browser operations (Browser Use).

Hermes originally had 12 browser tools — navigation, clicking, input, scrolling, screenshots — each handling its own task. When the Agent needed to fill out a form, it had to go step by step: click a field and call one tool, fill in content and call another tool, racking up over a dozen tool calls back and forth. That's how tokens got burned.

The new Browser Use mode replaces all 12 tools with a single one. Powered by @browser_use's CLI 3.0, the Agent no longer calls a separate tool for each click. Instead, it writes a script directly, bundling the entire browser operation into a single execution.

The results are straightforward. Nous's test data shows token usage dropped by 48% to 66%, with no loss in accuracy.

Why the savings are so significant. I've written plenty of long tasks where Agents need to interact with web pages, and the biggest takeaway is that every tool call carries overhead. Each time the Agent calls a tool, it has to re-understand the context — clicking requires back-and-forth confirmation, screenshots require image uploads. With 12 tools used in a scattered way, a huge chunk of tokens goes to that back-and-forth of "this step's done, what's next called?"

By writing it as a script, the Agent describes the entire flow in one go and lets CLI 3.0 execute it on its own, cutting out a lot of the round trips in between. What's saved is exactly the tokens that were previously wasted on tool switching.

For regular users, the most visible change is that the same tasks are faster and cheaper. Previously, a multi-step web operation would take several rounds; now a single script handles it. Spending half the tokens means the same budget can run more tasks, or the same tasks cost less.

The fact that accuracy didn't drop is also worth mentioning. In my experience, the more tool calls there are, the higher the chance of errors — each step requires re-aligning state. With scripting, the Agent completes operations within a continuous logic flow, which is actually more stable. Nous's test results back this up.

This update is a solid win for anyone currently using Hermes to get work done. Browser automation has shifted from "click one step, call once" to "write a block, run it once" — and the tokens saved are real money.

For the direction of Agent tools, I think this approach is also representative. When the number of tools balloons to a certain point, rather than keep adding more, it's better to teach the Agent to package operations into scripts. This could be the next direction for browser automation.

This update is highly useful — my recommendation is straightforward: Hermes update.