Benchmarks · composite

Arena Text · Coding

Crowdsourced pairwise human preference on open-ended text prompts (style control on); Bradley-Terry rating. Only the Coding category.

All results

215 rows

#	Model	Effort	Arena Text · Coding
1	Gemini 4 Carbon	high	1573 ±18
2	Opus 4.7	high	1551 ±6
3	Opus 5.5	high	1551 ±12
4	Fable 5	high	1550 ±6
5	Gemini 4 Argon	high	1550 ±12
6	Opus 4.6	high	1548 ±5
7	Opus 4.6	max	1546 ±5
8	Opus 4.7	max	1546 ±6
9	Muse Spark 1.2	xhigh	1543 ±11
10	GPT-6 Astra	max	1542 ±11
11	GPT-6.1 Sol	max	1541 ±14
12	Kimi K3	max	1540 ±7
13	MiMo V2.6 Pro	high	1538 ±18
14	Fable 5.1	max	1537 ±11
15	Opus 5	high	1534 ±6
16	Muse Spark 1.3	max	1534 ±9
17	Opus 4.8	high	1534 ±6
18	Muse Spark 1.1	max	1533 ±6
19	GPT-5.6 Sol	xhigh	1533 ±7
20	Opus 4.5	high	1530 ±7
21	Gemini 3.8 Flash	high	1530 ±7
22	Opus 4.8	max	1529 ±6
23	Sonnet 5.5	xhigh	1529 ±13
24	Muse Spark		1529 ±10
25	Opus 5	max	1528 ±7
26	Sonnet 4.6	max	1528 ±6
27	GLM 5.3 Flash	max	1526 ±7
28	Qwen 3.8 Max	max	1525 ±7
29	Qwen 3.7 Max	max	1523 ±18
30	DeepSeek V4.1 Flash	max	1523 ±10