Yolo Y. Tang

@yoloytang

Ph.D. Student @UofR | B.Eng. @SUSTechSZ. Working on LMMs/Agents for Video Understanding | Like: J-Pop/Cosplay/Drawing (@yoloytang_art) | She/Her/Hers

following
followers

DeepSeek Releases V4.1-Flash-1017, Claiming Improved Efficiency and Benchmark Gains

DeepSeek says the model is live on its website, app and API, and optimized for DeepSeekHarness and Tencent CodeBuddy.

143 Sources, 8h ago, updated 8h ago

BVB video-reconstruction leaderboard puts GPT-6 Astra xhigh first at 73.8

The BVB update lists GPT-6 Sol xhigh second at 71.4 and $0.40 per scene. The earlier announcement describes a video-understanding test in which agents reconstruct real videos in Blender.

2 Sources, 22d ago, first seen 22d ago

BVB: a video-understanding benchmark built around Blender reconstruction

The BVB announcement lists 288 real videos and 51 agent configurations, asking whether an AI agent that understands a video can reconstruct it.

1 Source, 36d ago, first seen 36d ago

Featured in

DeepSeek Releases V4.1-Flash-1017, Claiming Improved Efficiency and Benchmark Gains

BVB video-reconstruction leaderboard puts GPT-6 Astra xhigh first at 73.8

BVB: a video-understanding benchmark built around Blender reconstruction

Featured in

DeepSeek Releases V4.1-Flash-1017, Claiming Improved Efficiency and Benchmark Gains

BVB video-reconstruction leaderboard puts GPT-6 Astra xhigh first at 73.8

BVB: a video-understanding benchmark built around Blender reconstruction