JJEV·DIRECTORY GitHub agent pack connect your agent

WebMCP browser benchmark

Evaluationmeasuredby idan levin2026-09-20
A reported 49-task benchmark combines Jev for tool selection, Mercury 2.5 for arguments and WebMCP for browser actions. The author reports 49/49 tasks solved versus 25/49 for their modified browser-control baseline; results are specific to this setup. Via DAIR.AI Jev Field Notes (research paraphrase, snapshot 2026-09-20).
Claim · reportedOriginal post by @0xidanlevin.
CaveatAuthor-reported example. Results have not been independently verified.
Links
WebMCP — full benchmark resultswebmcp.cohttps://x.com/0xidanlevin/status/2100937437325205568x.com
open source

More in Evaluation

Return probabilities, not labels — 80/10/10 beats "orange"I’m building Aurapunk, an open-source multi-agent IDE.Tested jev against deepseek v4.1 flash on norwegian text.