Vals AI Launches Time Horizon Index, Pits GPT-5.6 Sol Against Kimi K3 in Five-Day Rocket Contest
Vals AI launched the Time Horizon Index to evaluate the long-horizon planning capabilities of artificial-intelligence agents. The benchmark's first contest features a livestreamed competition between GPT-5.6 Sol and Kimi K3 in the Kerbal Space Program simulation, where each model has five days to design rockets, write flight software, and complete 30 missions.
The event is being broadcast on X and Twitch, showing the models building a space program from scratch without pre-built rockets or coding shortcuts. The index aims to test whether autonomous systems can manage complex, sequential engineering tasks and handle design, launch, and landing phases independently.
From the sources (3 posts)
@valsaiCan an AI agent build, launch and land a rocket? Today we're releasing our Time Horizon Index and populating it with it’s first game, KSP. We give agents 5 days to design rockets, write flight software, and complete 30 missions. We're als
@valsaiWhat you're actually watching on the AI Space Race stream 🧵 Two frontier models are building a space program from scratch in Kerbal Space Program: GPT-5.6 Sol 🇺🇸 vs. Kimi K3 🇨🇳. No pre-built rockets, no cheats. Design, launch, land, repeat
@teortaxestexSol vs K3 compete in a long-horizon KSP run