gekro
GitHub LinkedIn
News

AI News

Google open-sources Harness2 agent framework; Voxel51 adds MCP tooling for multimodal data

Google released Harness2, a test-time agent refinement framework that improved GLM-5 by 11.5 points without weight updates; Voxel51 3.0 unified multimodal AI data platform with 45+ MCP tools.

1 min read 3 sources

Google Research released Harness2, an Apache-2.0 licensed framework that rewrites an agent’s harness at test time to improve performance. One refinement step using Harness2 lifted GLM-5 by 11.5 points on WorkBuddy-Bench with no weight updates, indicating a mechanism for post-hoc agent optimization without model retraining (MarkTechPost).

Voxel51 released version 3.0, unifying search, curation, annotation and evaluation of multimodal sensor data into a single platform. The update includes Voxel51 Agent with 45+ Model Context Protocol (MCP) tools, dataset versioning, and a free self-service tier, targeting physical AI teams working with sensor data (MarkTechPost). Separately, Perplexity open-sourced SPACE, a Firecracker microVM sandbox platform for AI agents under Apache 2.0, with each agent task running in isolated VMs and credentials controlled externally (MarkTechPost).

Compiled automatically from the linked sources and published without manual editing - a neutral summary of third-party reporting, for information only. Every claim links to its origin. Not original reporting.