MEDIC (clinical summarization)

Science

MEDIC's clinical summarization track, scored by rubric on coverage, conformity, consistency and conciseness and reported as an overall percentage. The overall figure is a mean of judged per-dimension rubric scores, not a pass rate over countable items, so no denominator is declared. Observed scores span a narrow high band (roughly 55 to 88), which is characteristic of rubric-judged generation and means the family carries less discrimination than its range suggests.

Top models (higher is better)

Model	Score
DeepSeek-V3	87.8
GPT-4.1 Mini	87.5
Qwen2 72B Instruct	87.5
Phi-4	87.4
O4 Mini	87.2
Mistral Large 2 (Instruct 2407)	87.1
Qwen3 32B	87.0
Qwen3-235B-A22B	86.8
Qwen3-30B-A3B	86.6
Qwen2.5 7B Instruct	86.2
DeepSeek-R1-Distill-Qwen-32B	86.2
Qwen3 14B	86.1
DeepSeek R1 Distill Llama 8B	85.8
GPT-4.1	85.8
DeepSeek R1 Distill Qwen 14B	85.8
DeepSeek-R1-Distill-Llama-70B	85.6
Qwen2.5 3B Instruct	85.6
Mistral 7B Instruct v0.3	85.5
Aya Expanse 32B	85.2
Qwen3 4B	85.0
QwQ-32B-Preview	84.8
Llama 3.1 Nemotron 70B Instruct HF	84.7
Llama 4 Scout Instruct	84.3
Llama 3.1 70B Instruct	84.3
Qwen2.5 72B	84.1
Llama 3.1 405B Instruct	84.0
Kimi K2 Thinking	83.8
Llama 3.1 8B Instruct	83.8
Qwen3 1.7B	83.7
Qwen3 8B	83.7
Llama 3.2 3B Instruct	83.5
Llama 3.3 70B Instruct	83.2
Llama 3 70B Instruct	83.1
DeepSeek-V3.1	83.0
Llama 4 Maverick Instruct	82.9
Qwen3 0.6B	82.4
DeepSeek-R1-Distill-Qwen-1.5B	82.3
gpt-oss-20b	82.1
Mistral Large 3 675B Instruct 2512	81.7
Phi-3.5 Mini Instruct	81.6