Optimizing the mannequin and agent collectively might enhance planning and context dealing with, however any aggressive benefit would must be demonstrated by means of higher outcomes on enterprise initiatives whereas decreasing the necessity for human intervention, mentioned Pareekh Jain, CEO of Pareekh Consulting.
Meta reported that Muse Spark 1.2 achieved an 82.9% move@1 rating on Terminal-Bench 2.1, behind Claude Opus 5 however barely forward of GPT-5.6 Terra. On DeepSWE 1.1, the mannequin scored 59.3%, trailing each rivals.
For Terminal-Bench 2.1 and DeepSWE 1.1, Meta evaluated every mannequin with its chosen coding agent relatively than utilizing the identical agent all through. It additionally acknowledged that rival proprietary fashions might have carried out otherwise beneath instruments and prompts designed particularly for them.
