Benchmarks · composite

Arena Text · English

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

All results

216 rows

#	Model	Effort	Arena Text · English
1	Gemini 4 Argon	high	1523 ±9
2	Gemini 4 Carbon	high	1523 ±15
3	Opus 5.5	high	1516 ±10
4	Opus 4.6	high	1513 ±4
5	Fable 5	high	1512 ±5
6	Opus 4.7	high	1510 ±5
7	Opus 4.6	max	1506 ±4
8	Fable 5.1	max	1502 ±8
9	Muse Spark 1.3	max	1501 ±7
10	Opus 4.7	max	1500 ±5
11	Gemini 3.8 Flash	high	1498 ±6
12	Muse Spark		1496 ±8
13	MiMo V2.6 Pro	high	1495 ±15
14	Kimi K3	max	1495 ±6
15	Opus 5	high	1494 ±5
16	Opus 5	max	1494 ±6
17	GPT-6.1 Sol	max	1493 ±11
18	Muse Spark 1.2	xhigh	1493 ±9
19	Qwen 3.8 Max	max	1492 ±6
20	Muse Spark 1.1	max	1491 ±5
21	GLM 5.3	max	1491 ±7
22	Opus 4.8	high	1490 ±5
23	Sonnet 5.5	xhigh	1490 ±10
24	Gemini 3.1 Pro	high	1490 ±4
25	GPT-5.6 Sol	xhigh	1489 ±6
26	DeepSeek V4.1 Flash	max	1488 ±9
27	Gemini 3.6 Flash	high	1487 ±6
28	Gemini 3 Pro	high	1487 ±5
29	Gemini 3.7 Flash	high	1486 ±6
30	GPT-6 Astra	max	1486 ±8