Benchmarks · composite

Arena Text · Longer queries

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

All results

214 rows

#	Model	Effort	Score
1	Gemini 4 Carbon	high	1542 ±15
2	Gemini 4 Argon	high	1539 ±9
3	Opus 5.5	high	1525 ±10
4	Fable 5	high	1522 ±6
5	Opus 4.6	high	1521 ±5
6	Opus 4.6	max	1516 ±5
7	Opus 4.7	high	1515 ±5
8	Fable 5.1	max	1512 ±8
9	Gemini 3.8 Flash	high	1512 ±6
10	Opus 4.7	max	1507 ±5
11	Opus 5	high	1503 ±5
12	Opus 4.8	high	1503 ±5
13	Kimi K3	max	1503 ±6
14	Opus 5	max	1502 ±6
15	Gemini 3.7 Flash	high	1501 ±6
16	Opus 4.8	max	1499 ±5
17	Sonnet 5.5	xhigh	1499 ±10
18	Gemini 3.1 Pro	high	1499 ±4
19	GPT-5.6 Sol	xhigh	1498 ±6
20	Muse Spark 1.3	max	1496 ±7
21	MiMo V2.6 Pro	high	1495 ±15
22	Opus 4.5	high	1495 ±7
23	Sonnet 4.6	max	1492 ±5
24	Qwen 3.8 Max	max	1492 ±6
25	GPT-6.1 Sol	max	1491 ±11
26	Gemini 3 Pro	high	1490 ±6
27	Qwen 3.7 Max	max	1490 ±15
28	Muse Spark 1.2	xhigh	1489 ±9
29	GPT-6 Astra	max	1488 ±8
30	GPT-5.5	high	1487 ±5