VIDRAFT's Darwin-180B-RSI scores 100% on AIME and HMMT 2026
Korean startup VIDRAFT says its open-weight Darwin-180B-RSI is the first model to hit 100% on AIME 2026 (30/30) and HMMT February 2026 (33/33) on Hugging Face's official leaderboards. The result used 16-sample majority voting with a 131K-token thinking budget.
- AIME 2026: 100% majority of 16, 98.75% mean over 16 samples
- HMMT Feb 2026: 100% majority and 96.59% mean over 16 samples
- At a 65K budget misses were truncated solutions, not wrong math
- Previous top scores: 97.1% on AIME and 92.7% on HMMT
Read next
AI