Benchmarks / Arena (formerly LMArena)

Reported by Arena (formerly LMArena)

Arena (formerly LMArena)

Human preference combined with labels of whether each answer's checkable claims were true (factuality weighted 25% by default).

Results dated
25 Sep 2026
Models
177
Unit
Arena rating
Licence
Creative Commons Attribution 4.0 International
Arena Text factuality: overall, Arena rating, higher is better
#ModelArena Text factuality: overall · 95% range
Arena rating, higher is better
1 Claude Fable 5.1Anthropic · Claude Fable 5.1 (max)
1,501
1,495–1,507
1 Claude Opus 4.6Anthropic · Claude Opus 4.6 (high)
1,494
1,491–1,497
2 Claude Opus 5Anthropic · Claude Opus 5 (max)
1,488
1,484–1,493
2 Muse Spark 1.2Meta
1,484
1,475–1,493
3 Claude Opus 4.6Anthropic
1,488
1,486–1,491
3 GPT-5.5OpenAI · GPT-5.5 (high)
1,486
1,483–1,489
3 GPT-5.4OpenAI · GPT-5.4 (high)
1,485
1,482–1,488
3 Claude Fable 5Anthropic · Claude Fable 5 (high)
1,483
1,480–1,487
3 MiMo-V2.6-ProXiaomi
1,483
1,475–1,491
3 qwen3.7-max-previewAlibaba
1,481
1,472–1,489
4 GPT-5.5OpenAI
1,482
1,480–1,485
4 gemini-3-proGoogle
1,481
1,478–1,485
4 Muse Spark 1.3Meta · Muse Spark 1.3 (max)
1,479
1,473–1,485
6 Gemini 3.7 FlashGoogle · Gemini 3.7 Flash (high)
1,477
1,473–1,482
6 Gemini 3.8 FlashGoogle · Gemini 3.8 Flash (high)
1,477
1,473–1,482
7 Claude Opus 5Anthropic · Claude Opus 5 (high)
1,477
1,474–1,480
7 Claude Opus 4.7Anthropic · Claude Opus 4.7 (high)
1,477
1,474–1,480
9 Gemini 3.5 FlashGoogle · Gemini 3.5 Flash (high)
1,476
1,472–1,479
9 GPT-5.6 SolOpenAI · GPT-5.6 Sol (xhigh)
1,476
1,472–1,479
9 GPT-5.4OpenAI
1,475
1,473–1,478
9 Claude Opus 4.7Anthropic
1,475
1,472–1,479
9 qwen3.5-max-previewAlibaba
1,475
1,470–1,479
10 Kimi K3Moonshot AI · Kimi K3 (max)
1,472
1,468–1,476
10 DeepSeek V4.1 FlashDeepSeek · DeepSeek V4.1 Flash (max)
1,471
1,464–1,478
10 Qwen3.8 Max (0902)Alibaba
1,471
1,466–1,475
13 Gemini 3.1 Pro PreviewGoogle
1,472
1,469–1,474
14 Gemini 3.6 FlashGoogle · Gemini 3.6 Flash (high)
1,470
1,466–1,474
15 GLM 5.3Z.ai · GLM 5.3 (max)
1,468
1,463–1,473
15 muse-sparkMeta
1,467
1,462–1,473
18 Claude Opus 4.8Anthropic · Claude Opus 4.8 (high)
1,469
1,466–1,472
22 gemini-3-flashGoogle
1,468
1,464–1,472
22 GLM 5.3 FlashZ.ai
1,467
1,462–1,471
22 GPT-6 AstraOpenAI · GPT-6 Astra (max)
1,465
1,458–1,471
23 GPT-5.6 TerraOpenAI · GPT-5.6 Terra (xhigh)
1,467
1,463–1,470
24 Gemini 3.5 FlashGoogle · Gemini 3.5 Flash (medium)
1,466
1,462–1,469
24 MiMo-V2.5-ProXiaomi
1,465
1,462–1,468
24 Grok 4.5xAI
1,465
1,462–1,469
25 Claude Opus 4.8Anthropic
1,465
1,461–1,468
25 Muse Spark 1.1Meta
1,463
1,459–1,467
25 MiMo-V2.6-FlashXiaomi
1,459
1,451–1,468
26 GLM 5.2Z.ai · GLM 5.2 (max)
1,463
1,459–1,466
28 Gemini 2.5 ProGoogle
1,462
1,459–1,465
28 GPT-5.6 LunaOpenAI · GPT-5.6 Luna (xhigh)
1,461
1,458–1,465
29 Claude Sonnet 4.6Anthropic
1,461
1,458–1,464
32 Claude Opus 4.5Anthropic
1,459
1,456–1,463
35 Claude Opus 4.5Anthropic
1,459
1,456–1,462
35 GPT-5.1OpenAI · GPT-5.1 (high)
1,459
1,455–1,462
38 grok-4.20-multi-agent-beta-0309xAI
1,457
1,454–1,460
38 GLM 5.1Z.ai
1,456
1,453–1,460
38 Kimi K2.6Moonshot AI
1,456
1,452–1,460
38 ernie-5.1Baidu
1,456
1,452–1,460
38 Qwen3.6 Max PreviewAlibaba
1,452
1,445–1,460
38 amazon-nova-experimental-chat-26-02-10Amazon
1,452
1,443–1,460
41 grok-4.20-beta-0309-reasoningxAI
1,456
1,453–1,459
41 DeepSeek V4 Pro 0813DeepSeek
1,453
1,448–1,459
43 DeepSeek V4 Pro 0423DeepSeek
1,455
1,452–1,458
44 grok-4.20-beta1xAI
1,453
1,448–1,457
44 Gemma 4 31BGoogle
1,450
1,443–1,457
44 DeepSeek V3.2 ExpDeepSeek
1,446
1,435–1,457
45 Qwen3.7 PlusAlibaba
1,453
1,449–1,456
46 Claude Sonnet 5Anthropic · Claude Sonnet 5 (high)
1,452
1,448–1,455
46 InklingThinkingmachines
1,452
1,448–1,455
46 GLM 5Z.ai
1,452
1,448–1,456
47 gemini-3-flashGoogle
1,451
1,449–1,454
47 gpt-5.2-chat-latest-20260210OpenAI
1,451
1,447–1,455
48 Grok 4.6xAI · Grok 4.6 (high)
1,449
1,445–1,454
49 Hy3Tencent
1,447
1,441–1,453
53 Gemini 3.5 Flash LiteGoogle
1,448
1,444–1,451
53 Gemma 4 26B A4BGoogle
1,445
1,438–1,452
54 Claude Sonnet 4.5Anthropic
1,449
1,446–1,451
54 deepseek-v4-pro-previewDeepSeek
1,447
1,444–1,451
54 GPT-5.4 MiniOpenAI · GPT-5.4 Mini (high)
1,447
1,444–1,450
54 GLM 4.6Z.ai
1,447
1,443–1,451
54 Claude Sonnet 4.5Anthropic
1,447
1,444–1,449
54 GLM 4.7Z.ai
1,444
1,438–1,450
55 ernie-5.0-0110Baidu
1,445
1,442–1,449
58 ernie-5.0-preview-1203Baidu
1,442
1,436–1,448
58 qwen3-max-previewAlibaba
1,442
1,435–1,448
58 ernie-5.0-preview-1022Baidu
1,440
1,431–1,448
60 Kimi K2.5Moonshot AI
1,445
1,442–1,448
60 GPT-5.1OpenAI
1,445
1,441–1,448
61 Qwen3.5 397B A17BAlibaba
1,445
1,442–1,447
61 chatgpt-4o-latest-20250326OpenAI
1,444
1,441–1,447
61 gpt-5.5-instantOpenAI
1,443
1,439–1,448
61 mimo-v2-proXiaomi
1,443
1,439–1,448
63 GPT-5.2OpenAI
1,442
1,440–1,445
63 GLM 5V TurboZ.ai
1,440
1,435–1,446
65 GPT-5.2OpenAI · GPT-5.2 (high)
1,442
1,438–1,445
69 Qwen3.8 27BAlibaba
1,439
1,434–1,443
72 DeepSeek V4 Flash 0423DeepSeek
1,439
1,436–1,443
72 MiMo-V2.5Xiaomi
1,439
1,435–1,442
74 DeepSeek V3.2 ExpDeepSeek
1,436
1,430–1,442
77 Qwen3.6 PlusAlibaba
1,438
1,434–1,441
77 mimo-v2-omniXiaomi
1,436
1,431–1,441
78 DeepSeek V3.2DeepSeek
1,437
1,434–1,440
80 Gemini 2.5 FlashGoogle
1,437
1,434–1,439
81 Claude Opus 4.1Anthropic
1,435
1,432–1,439
84 MiniMax M3MiniMax
1,434
1,431–1,437
84 grok-4.1-thinkingxAI
1,434
1,431–1,437
84 deepseek-v4-flash-previewDeepSeek
1,434
1,430–1,437
84 Mistral Large 3 2512Mistral
1,434
1,431–1,436
84 Claude Opus 4.1Anthropic
1,433
1,429–1,437
84 Gemini 3.1 Flash Lite PreviewGoogle
1,433
1,430–1,436
84 DeepSeek V3.2DeepSeek
1,433
1,429–1,436
84 nvidia-nemotron-3-ultra-550b-a55b-nvfp4NVIDIA
1,430
1,424–1,437
84 Mistral Medium 3.5Mistral
1,430
1,424–1,436
88 Kimi K2.5Moonshot AI
1,429
1,423–1,435
92 Qwen3 235B A22B Instruct 2507Alibaba
1,431
1,429–1,434
92 grok-4.1xAI
1,431
1,428–1,434
93 amazon-nova-experimental-chat-12-10Amazon
1,424
1,415–1,434
94 longcat-flash-chat-2602-expMeituan
1,429
1,425–1,433
94 Qwen3 VL 235B A22B InstructAlibaba
1,414
1,396–1,432
95 Mistral Medium 3.1Mistral
1,429
1,426–1,431
95 Qwen3.5-122B-A10BAlibaba
1,428
1,424–1,432
95 GPT-6 SolOpenAI · GPT-6 Sol (max)
1,424
1,416–1,432
100 Grok 4.3xAI
1,428
1,425–1,431
103 GPT-6 LunaOpenAI · GPT-6 Luna (max)
1,422
1,414–1,430
105 Inkling SmallThinkingmachines
1,425
1,421–1,429
105 gemini-2.5-flash-preview-09-2025Google
1,425
1,420–1,429
105 Grok 4.7xAI · Grok 4.7 (xhigh)
1,420
1,412–1,429
106 gpt-5-chatOpenAI
1,422
1,415–1,428
107 GPT-5OpenAI · GPT-5 (high)
1,421
1,414–1,428
108 kimi-k2-thinking-turboMoonshot AI
1,423
1,420–1,426
108 MiniMax M2.7MiniMax
1,422
1,419–1,425
108 mimo-v2-flashXiaomi
1,422
1,419–1,425
108 Qwen3 Next 80B A3B InstructAlibaba
1,419
1,412–1,425
109 grok-4-0709xAI
1,419
1,413–1,425
113 Qwen3.5-27BAlibaba
1,420
1,416–1,424
113 grok-4-fast-reasoningxAI
1,417
1,411–1,423
114 dola-seed-2.0-proBytedance
1,419
1,416–1,422
114 gpt-5.3-chat-latestOpenAI
1,419
1,415–1,422
114 GPT-5.4 NanoOpenAI · GPT-5.4 Nano (high)
1,418
1,415–1,421
117 Claude Haiku 4.5Anthropic
1,418
1,416–1,420
117 Step 3.5 FlashStepfun
1,417
1,414–1,420
117 Hy3 previewTencent
1,413
1,406–1,420
119 amazon-nova-experimental-chat-11-10Amazon
1,415
1,411–1,419
119 Qwen3.5-FlashAlibaba
1,414
1,411–1,417
119 Kimi K2 0905Moonshot AI
1,403
1,389–1,418
119 Qwen3 Coder 480B A35BAlibaba
1,401
1,383–1,419
120 grok-4-1-fast-reasoningxAI
1,413
1,410–1,416
120 minimax-m2.1-previewMiniMax
1,411
1,406–1,416
120 mimo-v2-flashXiaomi
1,410
1,404–1,416
127 GPT-4.1OpenAI
1,408
1,402–1,415
129 Solar Pro 4Upstage
1,409
1,403–1,414
133 Qwen3.5-35B-A3BAlibaba
1,408
1,404–1,411
133 o3OpenAI
1,405
1,398–1,411
133 amazon-nova-experimental-chat-26-01-10Amazon
1,403
1,394–1,411
136 gemini-2.5-flash-lite-preview-09-2025Google
1,407
1,404–1,411
137 muse-glimmerMeta
1,401
1,393–1,410
139 amazon-nova-experimental-chat-10-20Amazon
1,399
1,393–1,405
143 GLM 4.6VZ.ai
1,392
1,381–1,402
144 GLM 4.5 AirZ.ai
1,395
1,388–1,402
144 GPT-5 MiniOpenAI · GPT-5 Mini (high)
1,394
1,387–1,401
145 Nemotron 3 SuperNVIDIA
1,390
1,383–1,397
149 MiniMax M2.5MiniMax
1,386
1,382–1,389
149 gpt-oss-120bOpenAI
1,385
1,378–1,391
149 Gemma 3 27BGoogle
1,380
1,369–1,391
152 Nova 2 LiteAmazon
1,380
1,374–1,386
152 granite-4.2-30bIBM
1,377
1,368–1,386
156 intellect-3Primeintellect
1,373
1,365–1,381
156 MiniMax M1MiniMax
1,370
1,359–1,381
156 o4 MiniOpenAI
1,368
1,358–1,378
157 trinity-large-previewArcee Ai
1,373
1,369–1,377
157 Trinity Large ThinkingArcee Ai
1,371
1,367–1,375
157 GLM 4.7 FlashZ.ai
1,370
1,365–1,376
157 Mercury 2Inception
1,368
1,359–1,378
158 Command ACohere
1,363
1,357–1,370
161 Nemotron 3 Nano 30B A3BNVIDIA
1,362
1,357–1,367
161 MiniMax M2MiniMax
1,360
1,353–1,367
162 nvidia-nemotron-3.5-lightning-30b-a3b-nvfp4NVIDIA
1,360
1,354–1,366
171 Granite 4.2 8BIBM
1,343
1,333–1,352
171 olmo-3.1-32b-instructAi2
1,342
1,336–1,348
171 granite-4.2-3bIBM
1,327
1,317–1,337
173 olmo-3-32b-thinkAi2
1,319
1,312–1,327
173 granite-4.1-8bIBM
1,311
1,302–1,320
173 mercuryInception
1,305
1,292–1,317
175 olmo-3.1-32b-thinkAi2
1,300
1,294–1,307

Models share a rank when their ranges overlap. Results as published by Arena (formerly LMArena); we do not re-run them.

What it measures

Human preference combined with labels of whether each answer's checkable claims were true (factuality weighted 25% by default).

What it does not measure

Not an error rate: claims that cannot be checked on the web are skipped, and preference still carries most of the weight.

Source

Arena (formerly LMArena) · Methodology

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