AIVentureBeatabout 3 hours ago
DeepSeek V4 Flash tops leaderboards, fails agent tests
9 min read
DeepSeek's V4 Flash topped model leaderboards but completed just 53.8% of complex agent tasks in Composio testing. Across 240 runs with eight harnesses including Claude Code and Codex, only six of 30 workflows passed every test.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
