@deepseek_ai @baseten At max reasoning with the provider adapter harness, DeepSeek V4.1 Flash 1017 used 36% fewer reasoning tokens on its first action than the original, averaged across 25 public games. Across 55 semi-private games, it used 48.6% fewer reasoning tokens on the first action.
DeepSeek V4.1 Flash 1017 cuts reasoning tokens on ARC-AGI games
AIARC Prize says DeepSeek V4.1 Flash 1017, run on Baseten at max reasoning, used 36% fewer reasoning tokens on its first action than the original across 25 public games. On 55 semi-private games, the model used 48.6% fewer reasoning tokens on the first action.
Source details
ARC Prize · x.comPublished · added here
Original title: @deepseek_ai @baseten At max reasoning with the provider adapter harness, DeepSeek V4.1 Flash 1017 used 36% fewer reasoning tokens on its...