Incretin Therapeutics: Evidence Synthesis

micro bench v1 · 89 released tasks · updated August 16, 2026

The suite covers the incretin therapeutic class: pivotal trial results, cardiovascular and renal outcome evidence, label facts, and guideline positions for GLP-1 receptor agonists and dual agonists, anchored to readouts from 2024 through 2026. Tasks demand specific quantities and study design details, not summaries, and each carries one penalized misstatement criterion scored without compensation.

Ranking

#	model		score	95% CI	safety fails	unanimity
1		Claude Fable 5 Anthropic	83.8	79.1–88.1	10/89	91%
2		Grok 4.6 xAI	81.3	77.0–85.3	12/89	88%
3		Claude Opus 5 Anthropic	78.3	72.9–83.4	14/89	89%
4		GPT-5.6 Sol (max) OpenAI	77.9	73.3–82.2	13/89	89%
5		Kimi K3 Moonshot AI	77.8	73.0–82.4	14/89	89%
6		GPT-5.6 Sol (high) OpenAI	77.7	73.1–82.0	13/89	89%
7		Muse Spark Meta	68.4	62.8–73.9	19/89	86%
8		Gemini 3.6 Google	57.9	51.2–64.3	19/89	81%
9	TM	Inkling Thinking Machines	53.9	47.7–60.1	23/89	82%
10		Claude Sonnet 5 Anthropic	50.1	44.3–56.1	20/89	82%
11		MiniMax M3 MiniMax	35.5	29.6–41.6	34/89	83%
12		MAI Thinking Microsoft AI	33.0	26.8–39.3	27/89	85%
13		GLM 5.2 Zhipu	29.2	23.9–34.9	33/89	83%
14		Mistral Medium 3.5 Mistral	14.4	10.6–18.5	47/89	89%
15		Nemotron 3.5 Lightning NVIDIA	7.8	4.7–11.3	48/89	94%

Consensus rubric credit, one answer per task, tools disabled. Safety fails count tasks with a consensus-met penalty criterion. Unanimity is the share of criterion judgments on which all three grader families agreed before adjudication.

Task difficulty

Construction and grading follow the bench-wide protocol: five-family authoring, two-family confirmation before any removal, blind author-excluded panel grading with escalation, and a confidential holdout. Details are on the methodology page.