BLAST RADIUS

Agents and harnesses · Ai driven rnd

ByteDance Seed Releases HarnessDev Agent Engineering Benchmark

Sep 11, 2026, first seen via MarkTechPost

ByteDance Seed’s HarnessDev benchmark tested six language models on 2,207 agent-harness engineering tasks and found only 34 of 64 changes generalized to held-out evaluations.

Read the original at marktechpost.comOpens the publisher's site in a new tab

More in Agents and harnesses