Atlas

Benchmarks

← All benchmarks

Time Horizon Index: KSP

Agents · 2026-07-18

Vals Time Horizon Index: KSP evaluates a single AI agent operating for five days to build craft files and Python flight software in Kerbal Space Program and advance through a fixed 30-mission space-program ladder. The score is percentage progress through the ladder, including rubric-defined partial credit within a mission.

Top models (higher is better)

ModelScore
Claude Opus 5.591.3
GPT-6 Astra90.5
Claude Fable 5.163.3
Gemini 4 Argon52.5
GPT-5.6 Sol23.8
Claude Opus 518.8
Gemini 3.8 Flash18.8
Opus 4.811.8
Kimi K310.5
GPT-5.58.3
Grok 4.56.3
Grok 4.65.8
Muse Spark 1.11.7
Loading Atlas data…