Benchmarks / Arena (formerly LMArena)

Reported by Arena (formerly LMArena)

Arena (formerly LMArena)

Human preference on questions about an uploaded document.

Last updated 2 Oct 2026

Results dated
13 Sep 2026
Results
44 configurations of 38 models
Unit
Arena rating
Licence
Creative Commons Attribution 4.0 International

Arena (formerly LMArena) in brief

What it measures. Arena (formerly LMArena) is a third-party benchmark, shown here with credit. Measured here: Human preference on questions about an uploaded document.

Latest results. , 38 models ranked.

Top 3 on Overall (Arena rating, higher is better, each model at its best setting):

  1. Claude Fable 5 (Anthropic): 1,499
  2. Claude Opus 4.6 (high reasoning) (Anthropic): 1,495
  3. Claude Opus 4.7 (Anthropic): 1,494

The first 7 are too close to tell apart, so read the order among them as a tie.

How to cite. Arena (formerly LMArena), via Spring Prompt, results dated 13 September 2026. https://springprompt.com/benchmarks/lmarena These results belong to Arena (formerly LMArena), not to us. Licence: Creative Commons Attribution 4.0 International. Credit Arena (formerly LMArena) first.

Overall: GPT-5.6 Terra

Top 15 of 38 results · Arena rating, higher is better · lines show the 95% range · ≈ cannot be told apart from the leader. Choose a model to highlight it.Clear highlight

  1. 1≈ Claude Fable 5Anthropic 1,499
  2. 2≈ Claude Opus 4.6 (high reasoning)Anthropic 1,495
  3. 3≈ Claude Opus 4.7Anthropic 1,494
  4. 4≈ Claude Fable 5.1 (max reasoning)Anthropic 1,493
  5. 5≈ Claude Opus 5 (high reasoning)Anthropic 1,490
  6. 6≈ GPT-5.6 Sol (extra-high reasoning)OpenAI 1,489
  7. 7≈ Claude Opus 4.8 (high reasoning)Anthropic 1,486
  8. 8 GPT-5.5OpenAI 1,483
  9. 9 Muse Spark 1.1Meta 1,479
  10. 10 Claude Sonnet 4.6Anthropic 1,478
  11. 11 GPT-5.6 Terra (extra-high reasoning)OpenAI 1,477
  12. 12 Claude Sonnet 5 (high reasoning)Anthropic 1,476
  13. 13 GPT-6 Astra (max reasoning)OpenAI 1,473
  14. 14 GPT-5.4OpenAI 1,470
  15. 15 Muse Spark 1.3 (max reasoning)Meta 1,468

Full results

Arena Document: overall, Arena rating, higher is better
#ModelOverall · 95% range
Arena rating, higher is better
Price
$ per million tokens, in / out
1≈ Claude Fable 5Anthropic
1,499
1,491–1,507
$10 / $50
2≈ Claude Opus 4.6 (high reasoning)Anthropic · best of 2 settings
1,495
1,489–1,502
$5 / $25
3≈ Claude Opus 4.7Anthropic · best of 2 settings
1,494
1,488–1,501
$5 / $25
4≈ Claude Fable 5.1 (max reasoning)Anthropic
1,493
1,478–1,508
$10 / $50
5≈ Claude Opus 5 (high reasoning)Anthropic
1,490
1,483–1,497
$5 / $25
6≈ GPT-5.6 Sol (extra-high reasoning)OpenAI
1,489
1,480–1,498
$4 / $20
7≈ Claude Opus 4.8 (high reasoning)Anthropic · best of 2 settings
1,486
1,479–1,493
$5 / $25
8 GPT-5.5OpenAI · best of 2 settings
1,483
1,476–1,489
$5 / $30
9 Muse Spark 1.1Meta
1,479
1,471–1,488
$1.25 / $4.25
10 Claude Sonnet 4.6Anthropic
1,478
1,472–1,484
$3 / $15
11 GPT-5.6 Terra (extra-high reasoning)OpenAI
1,477
1,468–1,485
$2 / $12
12 Claude Sonnet 5 (high reasoning)Anthropic
1,476
1,468–1,483
$2 / $10
13 GPT-6 Astra (max reasoning)OpenAI
1,473
1,458–1,487
$10 / $50
14 GPT-5.4OpenAI
1,470
1,464–1,476
$2.50 / $15
15 Muse Spark 1.3 (max reasoning)Meta
1,468
1,450–1,486
$1.25 / $4.25
16 Muse SparkMeta
1,467
1,449–1,486
–
17 Claude Opus 4.5Anthropic
1,465
1,454–1,475
$5 / $25
18 Grok 4.5xAI
1,464
1,456–1,473
$2 / $6
19 GPT-5.6 Luna (extra-high reasoning)OpenAI
1,462
1,454–1,470
$0.20 / $1.20
20 Grok 4.6 (high reasoning)xAI
1,461
1,449–1,473
$2 / $6
20 Gemini 3.5 Flash (medium reasoning)Google · best of 2 settings
1,461
1,453–1,468
$1.50 / $9
22 GPT-5.5 InstantOpenAI
1,460
1,452–1,469
–
23 Gemini 3.1 Pro PreviewGoogle
1,459
1,454–1,464
$2 / $12
24 Gemini 3.6 Flash (high reasoning)Google
1,453
1,443–1,464
$1.50 / $7.50
24 Qwen3.7-PlusAlibaba
1,453
1,443–1,462
$0.32 / $1.28
26 Gemini 3 Pro PreviewGoogle
1,451
1,442–1,460
–
27 Kimi K2.6Moonshot AI
1,450
1,442–1,458
$0.95 / $4
28 Claude Sonnet 4.5Anthropic
1,449
1,443–1,455
$3 / $15
29 Gemma 4 31BGoogle
1,445
1,437–1,452
$0.15 / $0.40
30 Grok 4.20 Beta (0309, reasoning)xAI
1,439
1,432–1,447
–
30 Gemini 3 Flash PreviewGoogle
1,439
1,430–1,449
$0.50 / $3
32 Kimi K2.5 (reasoning on)Moonshot AI
1,436
1,429–1,443
$0.57 / $2.85
33 MiniMax-M3MiniMax
1,433
1,425–1,441
$0.30 / $1.20
34 Gemini 2.5 ProGoogle
1,431
1,424–1,437
$1.25 / $10
35 GPT-5.2OpenAI · best of 2 settings
1,425
1,419–1,431
$1.75 / $14
36 Claude Haiku 4.5Anthropic
1,419
1,413–1,425
$1 / $5
37 GPT-5.1OpenAI
1,416
1,407–1,425
$1.25 / $10
38 GLM-5V-TurboZ.ai
1,400
1,392–1,408
$1.20 / $4

Swipe the table sideways for more columns.

Source: Arena (formerly LMArena), via Spring Prompt, springprompt.com/benchmarks/lmarena. Test your own prompt on these models →

Ranks follow the score as shown, so equal numbers share a rank. ≈ marks results whose 95% range overlaps the leader's: they cannot be told apart from it. Each model is shown at its best setting; show every setting. Results as published by Arena (formerly LMArena); we do not re-run them.

Progress over time

Overall, plotted at each model's release date, higher is better. Each model counts once, at its best setting. Scores are today's, not launch day's; 1 model without a known release date is left off. Hover or tap a mark for its model.

Best score so far

Every model at its release date. The line moves each time a new release beats every earlier one, and each new best is named where there is room.

1,3501,4001,4501,5001,550 Apr 2025JulOct2026AprJulOct Gemini 2.5 ProClaude Sonnet 4.5Gemini 3 Pro PreviewClaude Opus 4.5Claude Opus 4.6Claude Fable 5

Best score by lab

Each lab's best so far, stepping whenever one of its releases beats its own earlier models. The 6 labs with the best scores today; 3 others not drawn.

AnthropicOpenAIMetaxAIGoogleAlibaba

1,4001,4501,500 Apr 2025JulOct2026AprJulOct AnthropicOpenAIMetaxAIGoogleAlibaba

What it measures

Human preference on questions about an uploaded document.

What it does not measure

Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Contains data from the Arena Leaderboard Dataset by Arena, licensed under CC BY 4.0. Licence: Creative Commons Attribution 4.0 International.