Atlas

Benchmarks

← All benchmarks

RadLE 2.0 Handover Readiness Index (RadLE-H)

Multimodal

RadLE-H combines the reliability and coverage of diagnoses made at autonomous-level confidence with successful deferral of likely errors. It measures whether a system's confidence can route cases between autonomous handling and specialist review.

Top models (higher is better)

ModelScore
Human Expert Baseline (RadLE 2.0)52.0
Muse Spark 1.148.5
Claude Fable 544.2
Grok 4.527.3
GPT-5.6 Sol Pro24.7
Qwen3.7-Plus22.1
MiniMax M316.7
Gemma 4 31B (RadLE 2.0 label)15.4
GLM 5V Turbo12.9
Lingshu-32B8.8
OctoMed-7B8.7
Llama 4 Maverick (RadLE 2.0 label)6.8
Gemini 3.1 Pro Preview6.1
Nemotron 3 Nano Omni 30B A3B2.4
MedGemma 1.5 4B IT2.0
Mistral Large 3 2512 (RadLE 2.0 label)1.4
InternVL3.5-8B1.1
Loading Atlas data…