Arena WebDev

Human preference between agentic front-end web development outputs; Bradley-Terry rating.

All results

123 rows

#	Model	Effort	Arena WebDev
1	Opus 5.5	max	1810 ±13
2	GPT-6 Astra	max	1773 ±10
3	Sonnet 5.5	xhigh	1770 ±11
4	Gemini 4 Carbon	high	1768 ±15
5	GPT-6.1 Sol	max	1751 ±11
6	Fable 5.1	max	1746 ±9
7	Sonnet 5.5	high	1711 ±9
8	Kimi K3.1	max	1695 ±13
9	Opus 5	max	1691 ±6
10	Gemini 4 Argon	high	1679 ±11
11	GPT-6 Sol	max	1678 ±9
12	Qwen 3.8 Max	max	1676 ±7
13	GPT-6.1 Terra	max	1673 ±15
14	DeepSeek V4.1 Flash	max	1665 ±8
15	Opus 5	high	1658 ±5
16	Kimi K3	max	1656 ±6
17	Muse Spark 1.3	max	1656 ±7
18	Grok 4.7	xhigh	1637 ±9
19	MiMo V2.6 Flash	high	1637 ±8
20	Hy4	high	1631 ±10
21	Qwen 3.8 Flash	max	1628 ±8
22	MiMo V2.6 Pro	high	1625 ±8
23	Fable 5	high	1624 ±6
24	GLM 5.3	max	1623 ±7
25	Muse Spark 1.3	xhigh	1620 ±7
26	Grok 4.6	high	1616 ±7
27	GPT-5.6 Sol	max	1616 ±6
28	GLM 5.3 Flash	max	1607 ±6
29	GLM 5.2	max	1606 ±6
30	Qwen 3.8 27B	max	1590 ±6