Unverified50% confidenceBenchmarkExact time
The research team evaluated MAGS on three categories of tasks: 100 CUDA kernels, 100 terminal scripts, and 20 robotic arm tasks, totaling 220 samples.
1
Sources
50%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
Unverified该实验使用175张训练图和132张测试图的小规模数据集61% similarUnverifiedDuring the evaluation phase, the system solved 105 out of 120 tasks.60% similarUnverifiedThe research team manually annotated 100 real recording cases, labeling 348 scoring units and 162 localization events across ten combined label types.59% similarUnverifiedDuring the training phase, the system passed 995 out of 1,000 tasks.59% similarUnverifiedCOCO 数据集包含约12万张训练图像、80个类别、超过150万个目标实例，由微软研究院发布58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/931241API
curl https://kongchang.com/api/v1/knowledge/claims/931241MCP
get_claim(id=931241)