Benchmarks · composite

Arena Text · Non-English

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

All results

223 rows

#	Model	Effort	Arena Text · Non-English
1	Gemini 4 Carbon	high	1517 ±12
2	Gemini 4 Argon	high	1514 ±7
3	Muse Wonder	max	1497 ±8
4	Fable 5.1	max	1495 ±6
5	Opus 5.5	high	1494 ±7
6	Fable 5	high	1493 ±5
7	Opus 4.7	high	1490 ±5
8	Opus 4.6	high	1488 ±4
9	Gemini 3.8 Flash	high	1488 ±5
10	Muse Spark 1.2	xhigh	1487 ±7
11	Muse Spark 1.3	max	1485 ±6
12	Gemini 3.7 Flash	high	1485 ±6
13	Opus 4.6	max	1483 ±4
14	Muse Spark 1.1	max	1483 ±5
15	Opus 4.7	max	1483 ±5
16	Opus 5	max	1480 ±6
17	Opus 5	high	1480 ±4
18	Gemini 3.1 Pro	high	1478 ±4
19	Gemini 3 Pro	high	1478 ±5
20	Muse Spark		1477 ±7
21	Kimi K3	max	1476 ±5
22	GPT-5.6 Sol	xhigh	1475 ±5
23	GPT-5.5	high	1473 ±5
24	Qwen 3.8 Max	max	1473 ±5
25	Gemini 3.6 Flash	high	1472 ±5
26	Kimi K3.1	max	1472 ±13
27	Gemini 3.5 Flash	high	1470 ±5
28	GPT-5.5 Instant		1470 ±6
29	Opus 4.8	high	1470 ±5
30	GPT-6.1 Sol	max	1469 ±8