Vibe Code Bench 1-100

Can models extend a working web application across a long sequence of dependent requests? Vals AI's overall score for Vibe Code Bench 1-100.

22 models · 22 results · updated 2026-10-17 · vals.ai ↗

More from this benchmark

1 published measures

Time per task26 results · lower is better

22 rows

#	Model	Effort	Vibe Code Bench 1-100	Subscription $ / task	API $ / task	Time	Harness
1	Opus 5.5	xhigh	30.4%	$0.67	$44	2.5 h	OpenHands
2	Opus 5	max	28.5%	$0.72	$41	1.4 h	OpenHands
3	Fable 5.1	max	28.0%	$23	$150	2.9 h	OpenHands
4	GPT-6 Astra	max	27.6%	$7.25	$85	3.4 h	OpenHands
5	GPT-5.6 Luna	max	22.6%	$0.46	$4.91	2.0 h	OpenHands
6	Gemini 4 Argon	high	21.2%	$6.65	$19	1.7 h	OpenHands
7	Muse Spark 1.3	max	20.5%	$2.80	$5.60	20 min	OpenHands
8	Haiku 5.5	max	20.4%	$0.18	$8.30	4.1 h	OpenHands
9	GPT-5.6 Sol	max	20.0%	$2.62	$47	2.6 h	OpenHands
10	GLM 5.3	max	20.0%	$0.86	$8.24	1.4 h	OpenHands
11	Gemini 3.8 Flash	high	18.8%	$0.77	$19	1.1 h	OpenHands
12	Kimi K3	max	18.2%	$0.73	$9.40	1.8 h	OpenHands
13	DeepSeek V4 Pro 0813	max	17.5%	$5.89	$5.89	38 min	OpenHands
14	DeepSeek V4.1 Flash	high	16.4%	$0.043	$0.69	28 min	OpenHands
15	GLM 5.3 Flash	max	16.0%	free	$0.46	39 min	OpenHands
16	GPT-5.6 Terra	max	14.8%	$1.84	$33	1.8 h	OpenHands
17	Grok 4.6	high	14.8%	$2.70	$26	2.2 h	OpenHands
18	Mistral Large 4	high	14.3%	$6.75	$6.75	60 min	OpenHands
19	Sonnet 5	max	13.8%	$1.23	$71	4.3 h	OpenHands
20	Qwen 3.8 Max	max	12.8%	$9.68	$9.68	1.6 h	OpenHands
21	MiniMax M3	max	9.2%	$0.51	$3.03	26 min	OpenHands
22	Inkling	max	7.3%	free	$5.00	50 min	OpenHands