Bedside Bench

Five hundred physician-authored clinical cases spanning ten suites of medical reasoning and safety, from differential diagnosis and workup through escalation and medication safety. Each case is graded by an LLM judge against criteria the authoring physicians wrote. The headline score is the macro-average of the ten suite means, so no single suite can carry a model.

Last updated

Main Insights

Results

Bedside Bench results, ranked by score (avg@3), highest first. Last updated .
RankModelProviderScore (avg@3)Cost (USD / task)Duration (seconds / task)
1Doximity Ask V7Doximity95.4%——
2Claude Opus 5Anthropic94.6%$0.1265
3Muse Spark OpenMeta94.3%$0.01779
4GPT-6.1 SolOpenAI93.9%$0.038—
5GPT-6 AstraOpenAI93.4%$0.1154
6Gemini 4 ArgonGoogle92.7%$0.02938
7Kimi K3.1 ProMoonshot AI91.5%$0.046—
8GPT-5.6 SolOpenAI91.1%$0.05042
9Kimi K3Moonshot AI89.7%$0.05383
10Ember-1Fireworks89.1%$0.04154
11GPT-6 SolOpenAI88.1%$0.01528
12GPT-6 LunaOpenAI87.2%$0.001433
13GLM 5.3Z.ai86.9%$0.02195
14Gemini 3.8 FlashGoogle86.5%$0.009210
15DeepSeek V4.1 Flash (1017)DeepSeek85.7%$0.006637
16Grok 4.6xAI83.3%$0.01239
17DeepSeek V4.1 FlashDeepSeek82.9%$0.0084102
18DeepSeek V4 Pro 0813DeepSeek80.6%$0.01663