HaliteChallenge/Halite | ProgramBench

@twosigma's first artificial intelligence programming challenge

275

Generated Behavioral Tests

99.6%

Best Score

Claude Opus 5.5 (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)

29 runs

#	Model	Score	Cost	Calls
1	Claude Opus 5.5 (xhigh)	99.6%	$9.00	134	trace →
2	Claude Fable 5.1 (xhigh)	99.6%	$27.28	217	trace →
3	GPT 6 Astra (max)	99.6%	$15.81	71	trace →
4	Muse Wonder Full	96.7%	TBA	300	trace →
5	GPT 5.5 (xhigh)	95.6%	$7.37	59	trace →
6	GPT-5.6 Sol (xhigh)	94.2%	$6.43	47	trace →
7	Muse Wonder (xhigh)	93.5%	$26.91	265	trace →
8	GPT 5.5 (high)	93.1%	$2.90	40	trace →
9	Claude Opus 5 (xhigh)	89.8%	$42.45	238	trace →
10	Gemini 3.7 Flash	89.8%	$2.11	139	trace →
11	Claude Opus 4.8 (xhigh)	84.4%	$14.45	106	trace →
12	Gemini 3.6 Flash	84.0%	$3.51	121	trace →
13	Claude Opus 4.6	80.4%	$15.08	222	trace →
14	GPT 5.5	80.4%	$1.06	17	trace →
15	Gemini 3.1 Pro	75.6%	$1.56	73	trace →
16	Muse Spark 1.1 (xhigh)	72.0%	$1.37	187	trace →
17	Claude Opus 4.7	71.6%	$1.00	44	trace →
18	GPT-5.6 Sol	66.9%	$0.70	13	trace →
19	GLM-5.2	58.2%	$17.22	152	trace →
20	Gemini 3 Flash	47.3%	$0.52	71	trace →
21	Claude Haiku 4.5	42.9%	$0.45	106	trace →
22	Muse Spark 1.3 (max)	24.7%	$8.46	326	trace →
23	Muse Spark 1.3 (xhigh)	24.0%	$1.79	274	trace →
24	Claude Opus 4.7 (xhigh)	18.5%	$3.13	72	trace →
25	Gemini 3.5 Flash	18.5%	$5.29	172	trace →
26	GPT 5.4	18.5%	$0.20	6	trace →
27	Claude Sonnet 4.6	14.2%	$38.69	494	trace →
28	GPT 5 mini	1.5%	$0.04	22	trace →
29	GPT 5.4 mini	0.0%	$0.02	9	trace →

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