jgm/pandoc | ProgramBench

Universal markup converter

5,214

Generated Behavioral Tests

40.1%

Best Score

GPT 6 Astra (max)

Results by Model

Hover for details · Click a model to open it · Drag to zoom the x-axis · Shift-drag to pan · The line marks the Pareto frontier (best score per cost)

30 runs

#	Model	Score	Cost	Calls
1	GPT 6 Astra (max)	40.1%	$17.98	75	trace →
2	Muse Wonder (xhigh)	32.1%	$26.31	252	trace →
3	Claude Opus 5 (xhigh)	29.1%	$143.77	521	trace →
4	Muse Spark 1.1 (xhigh)	18.0%	$0.95	181	trace →
5	Muse Spark 1.3 (xhigh)	15.1%	$1.61	215	trace →
6	GPT 5.5 (high)	14.7%	$2.59	24	trace →
7	Claude Opus 4.8 (xhigh)	14.6%	$59.55	320	trace →
8	GPT-5.6 Sol	12.3%	$1.30	16	trace →
9	Claude Opus 4.7 (xhigh)	10.7%	$2.53	83	trace →
10	Gemini 3.6 Flash	10.5%	$6.31	199	trace →
11	GPT 5.5	10.1%	$1.63	21	trace →
12	GLM-5.2	8.4%	$29.87	306	trace →
13	Gemini 3.5 Flash	7.8%	$5.10	177	trace →
14	Claude Sonnet 4.6	6.7%	$35.83	568	trace →
15	Gemini 3.1 Pro	6.3%	$1.33	95	trace →
16	Gemini 3 Flash	6.2%	$0.20	67	trace →
17	GPT 5.4	5.2%	$0.24	8	trace →
18	Claude Opus 4.7	4.7%	$1.04	46	trace →
19	Muse Wonder Full	4.5%	TBA	482	trace →
20	Claude Fable 5.1 (xhigh)	3.9%	$82.02	210	trace →
21	Muse Spark 1.2 (xhigh)	3.4%	$1.19	180	trace →
22	GPT 5.4 mini	0.6%	$0.05	19	trace →
23	GPT 5 mini	0.4%	$0.06	20	trace →
24	Muse Spark 1.3 (max)	0.2%	$6.91	341	trace →
25	GPT-5.6 Sol (xhigh)	0.1%	$8.72	48	trace →
26	Gemini 3.7 Flash	0.1%	$1.35	123	trace →
27	Claude Opus 4.6	0.1%	$17.70	339	trace →
28	Claude Haiku 4.5	0.1%	$0.90	105	trace →
29	Claude Opus 5.5 (xhigh)	0.0%	$57.74	275	trace →
30	GPT 5.5 (xhigh)	0.0%	$3.56	41	trace →

Click a row to replay how that model rebuilt this program, or the model name to open its full run