Wikiprompt

Benchmarks de LLMs

Da Wikiprompt, a enciclopedia livre de prompts

Rankings ao vivo de modelos de linguagem e IA generativa, agregados diariamente de avaliacoes abertas da comunidade. Cada tabela cita sua fonte, e como a Wikiprompt guarda um snapshot diario, as mudancas nos rankings ficam registradas.

Ultima atualizacao: 2026-10-13 · Este e o ranking AO VIVO; os snapshots datados preservam cada ranking passado. Historico →

Explorador interativo

Filtre por arena, organizacao ou licenca, busque modelos, ordene a tabela e leia o grafico de custo vs performance. As tabelas abaixo continuam sendo a referencia citavel.

Voce esta vendo
LMArena · Arena de texto

Elo da comunidade em batalhas diretas votadas na LMArena.

Precos e contexto: OpenRouter (matching por nome)

Arena de texto

Clique em linhas ou pontos do grafico para fixar modelos e comparar so esses.

Custo vs performance

Preco de saida em USD por 1M de tokens (escala log, da OpenRouter) vs Elo da comunidade (LMArena). Canto superior esquerdo = melhor valor.

Quadrante mais atraenteLinha de ParetoEscala:

Mostrando um subconjunto curado (top por organizacao). Escolha modelos ou marque Mostrar todos.

14501500$0.1$1$10$100Preco de saida, $ por 1M tokens (log)Elo da comunidadeWikipromptclaude-opus-5-maxclaude-opus-4-6-highclaude-opus-5-highgemini-3.8-flash-highclaude-opus-4-6claude-fable-5-highclaude-opus-4-7-highgemini-3.7-flash-highclaude-opus-4-7gemini-3.5-flash-highgemini-3.1-pro-previewgemini-3.6-flash-highmuse-spark-1.1kimi-k3-maxgpt-5.5-highglm-5.3-flashmimo-v2.5-prodeepseek-v4-progrok-4.20-beta-0309-reason…inklingminimax-m3mistral-large-3

Elo ao longo do tempo (top modelos de texto)

gemini-3.7-flash-highclaude-opus-5-highclaude-opus-4-7-highclaude-opus-4-6claude-opus-4-6-highclaude-opus-5-maxgemini-3.8-flash-highclaude-fable-5-high08-2308-2408-2508-2608-2708-2808-2908-3009-0109-0209-0309-0409-0509-0609-0709-0809-0909-1009-1109-1209-1309-1409-1509-1709-1809-1909-2009-2209-2709-2910-0210-0410-0510-0610-0710-0810-0910-1010-1110-1210-13
Posicao â–²ModeloOrganizacaoPontuacao EloVotosEntrada $/1MSaida $/1MContexto
4claude-opus-5-maxanthropic1,50730,569$5$251,000,000
5claude-opus-4-6-highanthropic1,50379,415$4$201,000,000
6claude-opus-5-highanthropic1,50362,934$5$251,000,000
7gemini-3.8-flash-highgoogle1,49931,437$0.75$3.751,048,576
8claude-opus-4-6anthropic1,49884,095$4$201,000,000
10claude-fable-5-highanthropic1,49140,026$10$501,000,000
11claude-opus-4-7-highanthropic1,49065,192$4$201,000,000
13gemini-3.7-flash-highgoogle1,48624,173$0.75$3.751,048,576
14qwen3.8-maxalibaba1,48325,378
15claude-opus-4-7anthropic1,48366,315$4$201,000,000
17gemini-3.5-flash-highgoogle1,48248,533$1.5$91,048,576
18gemini-3.1-pro-previewgoogle1,481124,675$2$121,048,576
19gemini-3-progoogle1,47942,078$2$12131,072
20gemini-3.6-flash-highgoogle1,47936,172$0.75$3.751,048,576
21muse-spark-1.1meta1,47939,847$1.25$4.251,048,576
24kimi-k3-maxmoonshot1,47630,351$0.73$101,048,576
26gpt-5.5-highopenai1,47269,437$5$301,050,000
29glm-5.3-flashzai1,47127,678$0.03$0.91,048,576
33ernie-5.1baidu1,46839,536
36mimo-v2.5-proxiaomi1,46573,197$0.44$0.871,050,000
48deepseek-v4-prodeepseek1,45157,657$0.96$1.911,048,576
50grok-4.20-beta-0309-reasoningxai1,45166,524$1.25$2.52,000,000
54dola-seed-2.0-probytedance1,44879,195
72inklingthinky1,44137,744$1$4.05524,288
90minimax-m3minimax1,43361,220$0.3$1.21,048,576
97mistral-large-3mistral1,42881,759$0.5$1.5262,144
101longcat-flash-chat-2602-expmeituan1,42729,306
131amazon-nova-experimental-chat-11-10amazon1,41526,059

Os precos sao pareados com o listing mais proximo da OpenRouter por nome normalizado; variantes de esforco compartilham o preco do modelo base.

Conclusoes
  • • Lider atual da arena de texto: claude-opus-5-max (anthropic, Elo 1507)
  • • Melhor custo-beneficio (Elo por dolar, abaixo de $2/1M output): glm-5.3-flash (Elo 1471, $0.90/1M)
  • • Modelo aberto mais forte: kimi-k3-max (Kimi K3 license, #24)
  • • Lider em coding (pass rate do Aider): gpt-5 (high) (88%)
Conteudo
  1. Arena de texto
  2. Arena de desenvolvimento web
  3. Arena de visao
  4. Arena de texto para imagem
  5. Arena de edicao de imagem
  6. Arena de texto para video
  7. Arena de edicao de video
  8. Arena de busca
  9. Arena de agentes
  10. Arena de documentos
  11. Precos e janelas de contexto
  12. Mudancas recentes
  13. Fontes e metodologia

Arena de texto25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
4claude-opus-5-maxanthropic1,507
5claude-opus-4-6-highanthropic1,503
6claude-opus-5-highanthropic1,503
7gemini-3.8-flash-highgoogle1,499
8claude-opus-4-6anthropic1,498
10claude-fable-5-highanthropic1,491
11claude-opus-4-7-highanthropic1,490
13gemini-3.7-flash-highgoogle1,486
14qwen3.8-maxalibaba1,483
15claude-opus-4-7anthropic1,483
17gemini-3.5-flash-highgoogle1,482
18gemini-3.1-pro-previewgoogle1,481
19gemini-3-progoogle1,479
20gemini-3.6-flash-highgoogle1,479
21muse-spark-1.1meta1,479
22gemini-3.5-flash-mediumgoogle1,477
24kimi-k3-maxmoonshot1,476
26gpt-5.5-highopenai1,472
29glm-5.3-flashzai1,471
30qwen3.5-max-previewalibaba1,470
31glm-5.2-maxzai1,470
32gpt-5.4-highopenai1,469
33ernie-5.1baidu1,468
34gpt-5.5openai1,467
35gemini-3-flashgoogle1,466

Arena de desenvolvimento web25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1claude-opus-5.5-maxanthropic1,813
2gpt-6-astra-maxopenai1,786
3claude-sonnet-5.5-xhighanthropic1,774
4gpt-6.1-sol-maxopenai1,755
5claude-fable-5.1-maxanthropic1,744
6claude-sonnet-5.5-highanthropic1,715
7claude-opus-5-maxanthropic1,691
8gpt-6-sol-maxopenai1,688
9gemini-4-argon-highgoogle1,678
10qwen3.8-max-0902alibaba1,674
11qwen3.8-maxalibaba1,672
12claude-opus-5-highanthropic1,657
13muse-spark-1.3-maxmeta1,657
14kimi-k3-maxmoonshot1,654
15grok-4.7-xhighxai1,639
16mimo-v2.6-flashxiaomi1,637
17qwen3.8-flash-nextalibaba1,633
18hy4-previewtencent1,632
19mimo-v2.6-proxiaomi1,629
20muse-spark-1.3 (xHigh)meta1,626
21claude-fable-5-highanthropic1,625
22glm-5.3-maxzai1,622
23deepseek-v4.1-flash-maxdeepseek1,619
24gpt-5.6-sol-xhigh (codex-harness)openai1,618
25grok-4.6-highxai1,617

Arena de visao25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1claude-fable-5-highanthropic1,324
2claude-opus-5.5-highanthropic1,321
3claude-opus-5-highanthropic1,319
4claude-fable-5.1-maxanthropic1,318
5claude-opus-4-7anthropic1,316
6gemini-3.7-flash-highgoogle1,316
7claude-opus-4-6-highanthropic1,316
8gemini-3.8-flash-highgoogle1,314
9qwen3.8-maxalibaba1,314
10claude-opus-4-6anthropic1,313
11claude-opus-4-7-highanthropic1,313
12gemini-3.5-flash-highgoogle1,310
13muse-spark-1.3-maxmeta1,309
14gemini-3.5-flash-mediumgoogle1,309
15muse-sparkmeta1,306
16gemini-3-progoogle1,305
17muse-spark-1.2 (xHigh)meta1,305
18gpt-5.4-highopenai1,303
19muse-spark-1.3 (xHigh)meta1,301
20gemini-3.6-flash-highgoogle1,298
21gpt-5.5openai1,297
22gemini-3.1-pro-previewgoogle1,296
23glm-5.3-flashzai1,296
24claude-opus-4-8-highanthropic1,294
25muse-spark-1.1meta1,293

Arena de texto para imagem25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1gpt-image-2.5-sunburstopenai1,425
2gpt-image-2.5-flareopenai1,398
3gpt-image-2 (medium)openai1,383
4grok-imagine-image-2.0 (canvas)xai1,334
5mai-image-2.6microsoft-ai1,332
6gemini-nano-banana-2.1google1,328
7reve-2.1reve1,302
8grok-imagine-image-2.0 (20260801)xai1,294
9muse-imagemeta1,274
10reve-2.0reve1,269
11gemini-3.1-flash-image (nano-banana-2) [web-search]google1,262
12seedream-5.0-probytedance1,255
13qwen-image-3.0-proalibaba1,255
14mai-image-2.5microsoft-ai1,254
15gemini-3.1-flash-lite-image (nano-banana-2-lite)google1,250
16gemini-3-pro-image-2k (nano-banana-pro)google1,248
17gpt-image-1.5-high-fidelityopenai1,237
18gemini-3-pro-image-preview (nano-banana-pro)google1,234
19qwen-image-2.1alibaba1,223
20ideogram-4.0-qualityideogram1,205
21qwen-image-2.0-pro-2026-06-22alibaba1,192
22uni-1.1-maxluma-ai1,189
23uni-1.1luma-ai1,184
24mai-image-2microsoft-ai1,184
25grok-imagine-imagexai1,170

Arena de edicao de imagem25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1gpt-image-2.5-sunburstopenai1,524
2gpt-image-2.5-flareopenai1,481
3gpt-image-2 (medium)openai1,462
4grok-imagine-image-2.0 (canvas)xai1,439
5mai-image-2.6microsoft-ai1,428
6gemini-nano-banana-2.1google1,428
7grok-imagine-image-2.0 (20260801)xai1,425
8muse-imagemeta1,403
9mai-image-2.5microsoft-ai1,401
10seedream-5.0-probytedance1,394
11grok-imagine-image-quality (20260519)xai1,391
12gemini-3-pro-image-2k (nano-banana-pro)google1,390
13chatgpt-image-latest-high-fidelity (20251216)openai1,389
14gemini-3.1-flash-image (nano-banana-2) [web-search]google1,387
15gemini-3-pro-image-preview (nano-banana-pro)google1,386
16reve-2.1reve1,374
17gpt-image-1.5-high-fidelityopenai1,370
18qwen-image-2.1alibaba1,369
19reve-2.0reve1,359
20ideogram-4.5ideogram1,347
21uni-1.1-maxluma-ai1,334
22grok-imagine-imagexai1,329
23uni-1.1luma-ai1,315
24gemini-3.1-flash-lite-image (nano-banana-2-lite)google1,314
25qwen-image-2.0-pro-2026-06-22alibaba1,303

Arena de texto para video25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1gemini-omni-1.1-flashgoogle1,516
2gemini-omni-flashgoogle1,513
3flux-3-videobfl1,493
4grok-imagine-video-1.5-agentxai1,492
5dreamina-seedance-2.0-720pbytedance1,479
6wan3.0alibaba1,476
7dreamina-seedance-2.5-720pbytedance1,474
8minimax-h3minimax1,460
9muse-videometa1,456
10happyhorse-1.0aorizon1,427
11sora-2-proopenai1,368
12veo-3.1-audiogoogle1,364
13veo-3.1-audio-1080pgoogle1,362
14veo-3.1-fast-audiogoogle1,362
15veo-3.1-fast-audio-1080pgoogle1,358
16veo-3-fast-audiogoogle1,348
17sora-2openai1,343
18grok-imagine-video-720pxai1,342
19veo-3-audiogoogle1,340
20wan2.7-t2vwan1,337
21wan2.6-t2valibaba1,327
22seedance-v1.5-probytedance1,256
23veo-3google1,254
24veo-3-fastgoogle1,248
25wan2.5-t2v-previewalibaba1,245

Arena de edicao de video10 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1wan3.0alibaba1,414
2dreamina-seedance-2.5-720pbytedance1,410
3minimax-h3minimax1,392
4gemini-omni-flashgoogle1,367
5dreamina-seedance-2.0-720pbytedance1,365
6happyhorse-1.0aorizon1,307
7grok-imagine-videoxai1,258
8kling-o3-prokling1,255
9kling-o1-prokling1,197
10runway-gen4-alephrunway1,182

Arena de agentes25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1Claude Opus 5.5 (Max)anthropic
2GPT 6.1 Sol (Max)openai
3GPT 6 Sol (Max)openai
4Gemini 4 Argon (High)google
5GPT 6 Astra (Max)openai
6Claude Fable 5.1 (Max)anthropic
7Claude Sonnet 5.5 (Max)anthropic
8Claude Fable 5 (High)anthropic
9Claude Opus 5 (Max)anthropic
10Claude Opus 4.8 (High)anthropic
11GPT 5.6 Sol (xHigh)openai
12Gemini 3.8 Flash (High)google
13GPT 5.5 (xHigh)openai
14Claude Sonnet 5 (High)anthropic
15Grok 4.7 (xHigh)xai
16Deepseek V4.1 Flash (Max)deepseek
17Muse Spark 1.3 (Max)meta
18MiMo V2.6 Proxiaomi
19GLM 5.2 (Max)zai
20Kimi K3 (Max)moonshot
21DeepSeek V4 Pro (High) (0813)deepseek
22GPT 6 Luna (Max)openai
23GPT 5.4 (High)openai
24Hy4 previewtencent
25MiMo V2.6 Flashxiaomi

Arena de documentos25 modelosRanking completo →

Fontes e metodologia: LMArena · Elo da comunidade em batalhas diretas votadas na LMArena.

PosicaoModeloOrganizacaoPontuacao Elo
1claude-opus-5-highanthropic1,516
2claude-fable-5.1-maxanthropic1,513
3claude-opus-4-6-highanthropic1,507
4claude-opus-4-6anthropic1,507
5claude-fable-5anthropic1,496
6claude-opus-4-7anthropic1,495
7claude-opus-4-7-highanthropic1,495
8gpt-5.5openai1,486
9gpt-5.5-highopenai1,484
10gpt-5.6-sol-xhighopenai1,483
11claude-sonnet-4-6anthropic1,482
12claude-opus-4-8-highanthropic1,475
13gpt-5.6-terra-xhighopenai1,472
14gpt-5.4openai1,471
15muse-spark-1.3-maxmeta1,471
16gpt-6-astra-maxopenai1,468
17claude-sonnet-5-highanthropic1,466
18muse-spark-1.1meta1,465
19claude-opus-4-8anthropic1,464
20gemini-3.5-flash-mediumgoogle1,463
21claude-opus-4-5-20251101anthropic1,462
22gemini-3.5-flash-highgoogle1,461
23gpt-5.6-luna-xhighopenai1,457
24gemini-3.6-flash-highgoogle1,456
25grok-4.6-highxai1,452

Outros leaderboards

LiveBench

Metrica: average score (0-100) · livebench.ai

PosicaoModeloOrganizacaoPontuacao
1claude-fable-5-1-max-effort83.8
2claude-fable-5-max-effort83.4
3claude-opus-5-5-max-effort83.2
4gpt-6-astra-max83.1
5gpt-6.1-sol-max82.5
6muse-spark-1.3-xhigh82.4
7claude-opus-5-5-xhigh-effort82.1
8muse-wonder-1.0-max81.9
9gpt-6.1-sol-xhigh81.8
10gpt-5.6-sol-max81.7
11deepseek-v4.1-flash-max81.5
12gpt-5.5-xhigh80.8
13claude-opus-5-max-effort80.5
14gemini-3.7-flash-high79.9
15gpt-6-sol-max79.8

Artificial Analysis (Intelligence Index)

Metrica: AA Intelligence Index (0-100) · artificialanalysis.ai

PosicaoModeloOrganizacaoPontuacao
1Claude Opus 5.5 (Max, Default Fallback)Anthropic57.6
2Claude Sonnet 5.5 (Max, Default Fallback)Anthropic56
3Claude Opus 5.5 (Xhigh, Default Fallback)Anthropic56
4Claude Opus 5.5 (High, Default Fallback)Anthropic53.6
5Claude Fable 5.1 (Max, Default Fallback)Anthropic53.4
6Claude Fable 5.1 (Xhigh, Default Fallback)Anthropic53.2
7GPT-6 Astra (Max)OpenAI52.7
8Gemini 4 Argon (High)Google52.6
9GPT-6 Astra (Xhigh)OpenAI52.4
10Claude Sonnet 5.5 (Xhigh, Default Fallback)Anthropic51.9
11GPT-6.1 Sol (Max)OpenAI51.8
12Claude Opus 5.5 (Medium, Default Fallback)Anthropic51.2
13Claude Fable 5.1 (High, Default Fallback)Anthropic51.2
14GPT-6.1 Sol (Xhigh)OpenAI51
15GPT-6 Astra (High)OpenAI50.9

Aider polyglot (coding)

Metrica: pass rate 2 (%) · aider.chat/docs/leaderboards/

PosicaoModeloOrganizacaoPontuacao
1gpt-5 (high)88
2gpt-5 (medium)86.7
3o3-pro (high)84.9
4gemini-2.5-pro-preview-06-05 (32k think)83.1
5o3 (high)81.3
6gpt-5 (low)81.3
7grok-4 (high)79.6
8gemini-2.5-pro-preview-06-05 (default think)79.1
9o3 (high) + gpt-4.178.2
10Gemini 2.5 Pro Preview 05-0676.9
11o376.9
12DeepSeek-V3.2-Exp (Reasoner)74.2
13Gemini 2.5 Pro Preview 03-2572.9
14o4-mini (high)72
15claude-opus-4-20250514 (32k thinking)72

BenchLM

Metrica: aggregated display score · benchlm.ai

PosicaoModeloOrganizacaoPontuacao
1Claude Opus 5.5Anthropic85.4
2GPT-6 AstraOpenAI83.3
3Claude Sonnet 5.5Anthropic82.5
4GPT-6.1 SolOpenAI81.6
5Claude Fable 5.1Anthropic81.2
6Gemini 4 ArgonGoogle80.6
7Claude Fable 5Anthropic78.7
8Claude Opus 5Anthropic78.7
9GPT-5.6 SolOpenAI77.1
10GPT-6 SolOpenAI74.3
11Muse Wonder FullMeta73.8
12MiMo-V2.6-ProXiaomi73.1
13GPT-5.5 ProOpenAI72.6
14Gemini 3.8 FlashGoogle72.4
15GPT-5.6 TerraOpenAI71.8

Precos e janelas de contexto

Fontes e metodologia: OpenRouter

Precos e metadata de contexto da API publica da OpenRouter. Precos em USD por 1M de tokens.

ModeloOrganizacaoContextoEntrada $/1MSaida $/1M
Auto Router (Beta)openrouter2,000,000$-1,000,000$-1,000,000
Pareto Code Routeropenrouter2,000,000$-1,000,000$-1,000,000
SpaceXAI: Grok 4.20 Multi-Agentx-ai2,000,000$1.25$2.5
SpaceXAI: Grok 4.20x-ai2,000,000$1.25$2.5
Auto Routeropenrouter2,000,000$-1,000,000$-1,000,000
Meta: Llama 4 Scoutmeta-llama1,310,720$0.18$0.59
OpenAI: GPT-6.1 Sol Proopenai1,050,000$2$10
OpenAI: GPT-6.1 Sol Pro (batch)openai1,050,000$1$5
OpenAI: GPT-6.1 Solopenai1,050,000$2$10
OpenAI: GPT-6.1 Sol (batch)openai1,050,000$1$5
OpenAI: GPT-6 Luna Proopenai1,050,000$0.1$0.5
OpenAI: GPT-6 Luna Pro (batch)openai1,050,000$0.05$0.25
OpenAI: GPT-6 Lunaopenai1,050,000$0.1$0.5
OpenAI: GPT-6 Luna (batch)openai1,050,000$0.05$0.25
OpenAI: GPT-6 Sol Proopenai1,050,000$2$10
OpenAI: GPT-6 Sol Pro (batch)openai1,050,000$1$5
OpenAI: GPT-6 Solopenai1,050,000$2$10
OpenAI: GPT-6 Sol (batch)openai1,050,000$1$5
Xiaomi: MiMo-V2.6-Flashxiaomi1,050,000$0.14$0.28
Xiaomi: MiMo-V2.6-Proxiaomi1,050,000$0.44$0.87

Mudancas recentes

Movimentos desde o snapshot anterior:

  • GPT 6.1 Sol (Max) â–² 5 → 2
  • GPT 6 Sol (Max) â–² 6 → 3
  • Gemini 4 Argon (High) â–² 7 → 4
  • GPT 6 Astra (Max) â–¼ 2 → 5
  • Claude Fable 5.1 (Max) â–¼ 3 → 6
  • Claude Sonnet 5.5 (Max) â–¼ 4 → 7
  • Claude Opus 4.8 (High) â–² 11 → 10
  • GPT 5.6 Sol (xHigh) â–² 12 → 11
  • Gemini 3.8 Flash (High) â–² 22 → 12
  • GPT 5.5 (xHigh) â–² 14 → 13
  • Claude Sonnet 5 (High) â–¼ 13 → 14
  • Grok 4.7 (xHigh) â–² 21 → 15

Fontes e metodologia

Os rankings sao agregados de avaliacoes abertas da comunidade. Pontuacoes Elo, intervalos de confianca e votos vem da LMArena (batalhas de modelos votadas pela comunidade, publicadas como dataset aberto). Precos e metadata de contexto vem da API publica da OpenRouter. A Wikiprompt tira um snapshot diario e preserva o historico; nao rodamos avaliacoes proprias.

  1. LMArena leaderboard dataset (Hugging Face)
  2. LMArena leaderboard
  3. OpenRouter models API
  4. Artificial Analysis (intelligence, speed and latency data)

Veja tambem