Clef-Flash 9B
A 9-billion-parameter local decision model designed for browser automation tasks. It drives action decisions by extracting webpage text content in real time, and is not responsible for text generation. The model is positioned as a lightweight, edge-side agent decision component.
Timeline (last 90 days)
A developer shared on Reddit a project using a decision model called Clef-Flash 9B to achieve automated web browsing running locally.
Clef-Flash 9B is a 9B-parameter decision model that runs locally and is specifically responsible for web browsing decisions.
The Clef-Flash 9B decision model cannot autonomously generate text; the system extracts and processes text from web pages during operation to support model decision-making.
In the demo, the task of querying when Tibo's usage quota resets took 17 seconds.
In the demo, the task of finding and downloading the latest stable version of Godot standard edition (non-.NET version) for Windows x86_64 took 13 seconds.
In the demo, the task of playing a YouTube video about the Jev model took 11 seconds.
In this demo the instructions were entered manually, and the author stated that connecting an LLM to issue instructions would form an end-to-end fully automated closed loop.
The system achieves a high success rate on websites with clear navigation structures, but its success rate drops significantly on websites containing complex UI elements.
Toggles and multiple tabs are two categories of difficulty for this system, because their states are difficult to read from page text.
The pure text extraction approach is lightweight and privacy-friendly, requiring no screenshots or visual models, but cannot capture visual state information—this is the trade-off between it and visual understanding solutions.
All Facts (10)
The pure text extraction approach is lightweight and privacy-friendly, requiring no screenshots or visual models, but cannot capture visual state information—this is the trade-off between it and visual understanding solutions.
50%UnverifiedClef-Flash 9B is a 9B-parameter decision model that runs locally and is specifically responsible for web browsing decisions.
50%UnverifiedThe Clef-Flash 9B decision model cannot autonomously generate text; the system extracts and processes text from web pages during operation to support model decision-making.
50%UnverifiedIn the demo, the task of querying when Tibo's usage quota resets took 17 seconds.
50%UnverifiedIn the demo, the task of finding and downloading the latest stable version of Godot standard edition (non-.NET version) for Windows x86_64 took 13 seconds.
50%UnverifiedIn the demo, the task of playing a YouTube video about the Jev model took 11 seconds.
50%UnverifiedIn this demo the instructions were entered manually, and the author stated that connecting an LLM to issue instructions would form an end-to-end fully automated closed loop.
50%UnverifiedThe system achieves a high success rate on websites with clear navigation structures, but its success rate drops significantly on websites containing complex UI elements.
50%UnverifiedToggles and multiple tabs are two categories of difficulty for this system, because their states are difficult to read from page text.
50%UnverifiedA developer shared on Reddit a project using a decision model called Clef-Flash 9B to achieve automated web browsing running locally.
50%