Inference
EN
Vals AI

Time Horizon Index: KSP

Can an AI agent build and run a space program in Kerbal Space Program?

ModelsModels with a result.13
Top resultBest result on this benchmark.91.3%Claude Opus 5.5
Top-3 spreadPoints from first to third.28.0 pts
UpdatedDate the source changed the results.7 Oct 2026

Result and cost

13 of 13 results

Results

13 results
#ModelResultThe score from the source.CostUS dollars to run one task.TimeTime to run one task.
01Claude Opus 5.591.3%$921.760.0 s
02GPT-6 Astra90.5%$4,206.590.0 s
03Claude Fable 5.163.3%$4,655.010.0 s
04Gemini 4 Argon52.5% ±8.9$708.630.0 s
05GPT-5.6 Sol23.8%$2,027.200.0 s
06Claude Opus 518.8%$1,348.770.0 s
06Gemini 3.8 Flash18.8%$543.140.0 s
08Claude Opus 4.811.8%$1,499.320.0 s
09Kimi K310.5% ±5.4$216.430.0 s
10GPT-5.58.3%$516.360.0 s
11Grok 4.56.3%$344.040.0 s
12Grok 4.65.8%$225.120.0 s
13Muse Spark 1.11.7%$144.920.0 s