Benchmark Almanac

Kuaishou

We track 4 benchmarks that Kuaishou made or helped make. Its models are on its lab page.

2 in use 2 with public code 0 with a leaderboard here

Benchmarks

Name Kind Released Best score Status
KAT Code Bench Agents turn a natural-language requirement into a patch in a real repository drawn from Kuaishou internal development work. Each task pins its commit, environment and an executable verification entry point across 12 languages. Fixing real repositories Jul 2026 53.1% In use
KAT Claw Bench Agents complete business tool-use tasks from Kuaishou internal agent needs in seven categories, such as office work, content operations, data analysis and monitoring. Automatic checks grade objective outputs, and model and human review grade open deliverables. General assistants Jul 2026 85.5% In use
PlayWorld An agent player pursues objectives inside video world models. Gemini 3.1 Pro applies weighted rubrics for geometry, interaction, and state evolution across the resulting videos. Interactive world models Aug 2026 Not in cards
VideoGen-RewardBench Models rank two generated videos for the same prompt. The scorer compares their rankings with human preferences for visual quality, motion, prompt alignment, and overall quality. Video understanding Jan 2025 Not in cards