Unverified50% confidenceBenchmarkExact time
The team has performed post-training of Cosmos on the ZOE dataset, achieving SOTA-level grasp-and-place policy results.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/3/2026
First Seen
Valid until: 1/1/2027
Sources
Related Entities
Related Claims
UnverifiedThe team publicly released the trained model weights, training data, training and inference code, submitted solutions, and an evaluation benchmark.67% similarUnverifiedThe training incorporated behavioral metrics such as merge rate, code review time, and repository size, ultimately generating more than 1 million step-to-step state transition data entries.64% similarUnverifiedMethods such as freezing LLM backbone parameters during the visual alignment training phase, staged training, data ratio replay, and LoRA can mitigate catastrophic forgetting in multimodal training.62% similarUnverifiedSWE-bench上已有多个研究团队记录到「patch记忆」现象，即模型通过识别问题描述特征检索训练时见过的补丁61% similarUnverifiedThe core value of experiment tracking platforms lies in extending the reproducibility of model training from the code layer to the experiment context layer.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/969722API
curl https://kongchang.com/api/v1/knowledge/claims/969722MCP
get_claim(id=969722)