Benchmarks · composite

Arena Text · Expert

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

All results

213 rows

#	Model	Effort	Arena Text · Expert
1	Gemini 4 Carbon	high	1581 ±28
2	Fable 5	high	1550 ±9
3	Opus 4.6	high	1544 ±8
4	Gemini 4 Argon	high	1541 ±17
5	Opus 5.5	high	1538 ±18
6	Opus 5	high	1537 ±8
7	Opus 4.7	high	1534 ±8
8	Opus 4.6	max	1533 ±7
9	Opus 4.7	max	1533 ±8
10	Fable 5.1	max	1533 ±16
11	Sonnet 5.5	xhigh	1532 ±19
12	Kimi K3	max	1532 ±11
13	GPT-5.6 Sol	xhigh	1532 ±9
14	MiMo V2.6 Pro	high	1531 ±27
15	Opus 5	max	1530 ±11
16	Gemini 3.8 Flash	high	1527 ±9
17	Opus 4.8	high	1527 ±8
18	GPT-6.1 Sol	max	1526 ±20
19	Muse Spark 1.3	max	1522 ±12
20	GLM 5.3	max	1521 ±12
21	Gemini 3.7 Flash	high	1518 ±10
22	GPT-6 Astra	max	1517 ±16
23	DeepSeek V4.1 Flash	max	1516 ±15
24	Opus 4.8	max	1515 ±8
25	GPT-5.4	high	1514 ±8
26	Qwen 3.8 Max	max	1513 ±10
27	GLM 5.3 Flash	max	1513 ±10
28	GPT-5.5	high	1512 ±8
29	Sonnet 5	high	1511 ±8
30	GPT-5.5	xhigh	1510 ±8