Starblink is a shortened version of our prototype intelligence test and AI benchmark Starburst, wherein one tries to figure out the laws of physics in a simple fictional universe. The full Starburst leaderboard is available at:
We are compiling and publishing the Starblink leaderboard to determine the inter-task reliability of Starburst-like tasks for LLMs and people, to provide further discernment, and to provide more robust measurements of model capability.
If you wish to try Starblink or Starburst, email me at chapinalc@gmail.com.
Era 1
1 Human
Era 2
Era 3
GPT-5.2 (extended thinking) (3/5)
Era 4
Gemini 3 Pro (2/5)
Era 5
Grok 4 (3/5)
o3 (3/5)
Gemini-2.5-Pro (3/5)
o4-mini-high (2/5)
Era 6
Era 7
Era 8
Era 9
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.