Benchmarks · long-context

Arena Document

Human preference on document analysis and long-content reasoning; Bradley-Terry rating.

37 models · 43 results · data as of 2026-10-29 · arena.ai ↗

All results

43 rows

#	Model	Effort	Arena Document
1	Opus 5	high	1516 ±7
2	Fable 5.1	max	1513 ±15
3	Opus 4.6	high	1507 ±6
4	Opus 4.6	max	1507 ±6
5	Fable 5	max	1496 ±8
6	Opus 4.7	max	1495 ±6
7	Opus 4.7	high	1495 ±7
8	GPT-5.5	xhigh	1486 ±6
9	GPT-5.5	high	1484 ±6
10	GPT-5.6 Sol	xhigh	1483 ±9
11	Sonnet 4.6	max	1482 ±6
12	Opus 4.8	high	1475 ±7
13	GPT-5.6 Terra	xhigh	1472 ±8
14	GPT-5.4	xhigh	1471 ±6
15	Muse Spark 1.3	max	1471 ±18
16	GPT-6 Astra	max	1468 ±14
17	Sonnet 5	high	1466 ±8
18	Muse Spark 1.1	max	1465 ±9
19	Opus 4.8	max	1464 ±7
20	Gemini 3.5 Flash	medium	1463 ±8
21	Opus 4.5	high	1462 ±10
22	Gemini 3.5 Flash	high	1461 ±9
23	GPT-5.6 Luna	xhigh	1457 ±8
24	Gemini 3.6 Flash	high	1456 ±11
25	Grok 4.6	high	1452 ±12
26	Grok 4.5	max	1452 ±9
27	Kimi K2.6	max	1451 ±8
28	Sonnet 4.5	high	1450 ±6
29	Qwen 3.7 Plus	high	1444 ±9
30	Gemini 3.1 Pro	high	1444 ±5