Benchmarks · composite

Arena Text · Hard prompts

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

All results

216 rows

#	Model	Effort	Score
1	Gemini 4 Carbon	high	1548 ±12
2	Gemini 4 Argon	high	1544 ±8
3	Opus 5.5	high	1535 ±8
4	Opus 4.6	high	1531 ±4
5	Fable 5	high	1529 ±5
6	Opus 4.6	max	1526 ±4
7	Opus 4.7	high	1525 ±4
8	Fable 5.1	max	1524 ±7
9	Opus 4.7	max	1519 ±4
10	Gemini 3.8 Flash	high	1517 ±5
11	Kimi K3	max	1517 ±5
12	Opus 5	high	1516 ±4
13	Muse Spark 1.3	max	1516 ±6
14	Opus 5	max	1515 ±5
15	Muse Spark 1.2	xhigh	1513 ±8
16	Opus 4.8	high	1513 ±5
17	GPT-5.6 Sol	xhigh	1511 ±5
18	Muse Spark 1.1	max	1511 ±5
19	Gemini 3.7 Flash	high	1509 ±5
20	Gemini 3.1 Pro	high	1508 ±4
21	GPT-6.1 Sol	max	1506 ±9
22	Muse Spark		1506 ±7
23	Opus 4.8	max	1505 ±5
24	Sonnet 5.5	xhigh	1505 ±8
25	GLM 5.3	max	1505 ±6
26	MiMo V2.6 Pro	high	1504 ±12
27	Qwen 3.8 Max	max	1504 ±5
28	Sonnet 4.6	max	1503 ±4
29	Gemini 3 Pro	high	1502 ±5
30	GLM 5.3 Flash	max	1502 ±5