HN
Ranked #1 on Hacker News with 130 points and 76 comments.
At max effort, DeepSeek V4 Flash 0731 scores 89.0% on ARC-AGI-1 Semi-Private at $0.02 per task and 61.4% on ARC-AGI-2 Semi-Private at $0.04 per task.
Pass/fail per reasoning level across each benchmark.