Atlas

Benchmarks

← All benchmarks

RT-SAFE Success Rate

Safety · 2026-10-07

Share of RT-SAFE's 108 real-time navigation episodes (36 routes at easy, medium and hard actor density, provider-default reasoning, one simulation seed) in which the agent reaches its destination, regardless of collisions or other safety events along the way. It measures navigation completion rather than safety, is near ceiling for most models, and is kept as a diagnostic.

Top models (higher is better)

ModelScore
GPT-5.6 Sol98.1
GPT-6 Astra97.2
Claude Fable 5.196.3
Gemini 3.8 Flash94.4
Inkling94.4
Grok 4.692.6
DeepSeek V4 Flash Vision Exp91.7
Sonnet 590.7
Loading Atlas data…