Benchmarks · composite

Arena Text · Hard prompts (English)

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

All results

223 rows

#	Model	Effort	Score
1	Gemini 4 Carbon	high	1544 ±20
2	Gemini 4 Argon	high	1544 ±10
3	Muse Wonder	max	1537 ±12
4	Opus 4.6	high	1535 ±5
5	Opus 5.5	high	1533 ±11
6	Fable 5	high	1532 ±6
7	Opus 4.6	max	1530 ±5
8	Opus 4.7	high	1528 ±6
9	Opus 4.7	max	1521 ±6
10	Kimi K3.1	max	1520 ±23
11	Fable 5.1	max	1519 ±9
12	Gemini 3.8 Flash	high	1517 ±6
13	Opus 4.8	high	1517 ±6
14	Muse Spark 1.3	max	1516 ±8
15	GLM 5.3	max	1515 ±8
16	GPT-6.1 Sol	max	1515 ±12
17	Kimi K3	max	1515 ±7
18	Opus 5	high	1514 ±6
19	Opus 5	max	1512 ±7
20	Qwen 3.8 Max	max	1511 ±7
21	GPT-5.6 Sol	xhigh	1511 ±7
22	Muse Spark 1.2	xhigh	1511 ±10
23	GPT-6 Astra	max	1510 ±10
24	MiMo V2.6 Pro	max	1510 ±19
25	Muse Spark		1508 ±9
26	Gemini 3.1 Pro	high	1507 ±4
27	Sonnet 4.6	max	1506 ±5
28	Opus 4.8	max	1506 ±6
29	Sonnet 5.5	xhigh	1506 ±11
30	Muse Spark 1.1	max	1505 ±6