ariga/atlas | ProgramBench

Declarative schema migrations with schema-as-code workflows

1,318

Generated Behavioral Tests

91.4%

Best Score

Muse Wonder (xhigh)

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	Muse Wonder (xhigh)	91.4%	$54.24	464	trace →
2	GPT 6 Astra (max)	91.3%	$10.94	38	trace →
3	Claude Opus 5.5 (xhigh)	91.1%	$69.28	479	trace →
4	Muse Wonder Full	89.8%	TBA	492	trace →
5	Claude Fable 5.1 (xhigh)	82.9%	$35.22	151	trace →
6	Claude Opus 5 (xhigh)	82.5%	$46.17	242	trace →
7	Muse Spark 1.3 (max)	80.7%	$5.44	324	trace →
8	Muse Spark 1.3 (xhigh)	74.3%	$2.24	332	trace →
9	GPT 5.5 (xhigh)	72.5%	$9.62	80	trace →
10	GLM-5.2	62.5%	$16.91	190	trace →
11	Gemini 3.6 Flash	57.7%	$3.97	176	trace →
12	Claude Opus 4.8 (xhigh)	57.6%	$8.92	95	trace →
13	Claude Opus 4.7 (xhigh)	55.4%	$11.04	163	trace →
14	Claude Opus 4.7	54.8%	$8.11	173	trace →
15	GPT 5.5 (high)	54.6%	$3.23	28	trace →
16	Muse Spark 1.2 (xhigh)	52.2%	$0.12	168	trace →
17	GPT-5.6 Sol	47.2%	$1.15	18	trace →
18	GPT 5.5	45.7%	$1.58	31	trace →
19	Gemini 3.5 Flash	32.3%	$4.30	150	trace →
20	Claude Haiku 4.5	28.4%	$0.61	127	trace →
21	GPT 5.4	26.3%	$0.43	16	trace →
22	Gemini 3 Flash	22.5%	$0.27	104	trace →
23	Gemini 3.1 Pro	21.6%	$1.62	114	trace →
24	GPT 5 mini	20.4%	$0.02	15	trace →
25	GPT-5.6 Sol (xhigh)	9.3%	$5.49	34	trace →
26	Gemini 3.7 Flash	9.3%	$1.30	152	trace →
27	Claude Sonnet 4.6	9.3%	$35.07	626	trace →
28	Muse Spark 1.1 (xhigh)	9.2%	$0.12	136	trace →
29	Claude Opus 4.6	0.0%	$5.00	296	trace →
30	GPT 5.4 mini	0.0%	$0.06	18	trace →

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