Benchmarks · composite

Arena Text · Chinese

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

All results

212 rows

#	Model	Effort	Arena Text · Chinese
1	Gemini 4 Carbon	high	1588 ±28
2	Gemini 4 Argon	high	1579 ±20
3	Opus 5.5	high	1574 ±19
4	Fable 5.1	max	1565 ±17
5	Fable 5	high	1555 ±11
6	Opus 4.6	high	1550 ±9
7	Opus 5	high	1549 ±9
8	Opus 5	max	1547 ±12
9	Gemini 3.7 Flash	high	1547 ±12
10	Sonnet 5.5	xhigh	1545 ±21
11	Opus 4.6	max	1544 ±8
12	Kimi K3	max	1543 ±12
13	GPT-5.6 Sol	xhigh	1540 ±11
14	Qwen 3.8 Max	max	1538 ±13
15	GLM 5.3	max	1538 ±14
16	Opus 4.7	high	1537 ±10
17	Gemini 3.8 Flash	high	1536 ±10
18	Opus 4.7	max	1536 ±10
19	GLM 5.3 Flash	max	1530 ±11
20	Gemini 3.6 Flash	high	1530 ±12
21	Muse Spark 1.3	max	1529 ±15
22	Muse Spark 1.1	max	1529 ±11
23	Gemini 3.1 Pro	high	1528 ±7
24	GPT-5.5	xhigh	1526 ±10
25	Opus 4.8	high	1526 ±10
26	Opus 4.8	max	1525 ±10
27	Muse Spark		1525 ±20
28	GLM 5.2	max	1525 ±10
29	DeepSeek V4.1 Flash	max	1523 ±19
30	Gemini 3 Pro	high	1521 ±10