Intel / article

Mercury 2 On Pinchbench

Source: inceptionlabs.ai
Author: Inception
Date: Published Mar 24, 2026

Context

Inception reports a 78% task-success rate for Mercury 2 on PinchBench, which evaluates a model operating an through multistep workflows. The examples include scheduling, research, file management and coding, with success concerned with completing actions rather than merely describing them. This gives the result a specific meaning: it is evidence about Mercury 2's performance within the tested agent setup. The reported score offers a starting point for assessing task completion, while the article's separate speed claims require their own comparison conditions. Evaluation account

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.

Read the source www.inceptionlabs.ai