Todos os testes de desempenho

Avaliações de desempenho de roteamento de modelos

Compare o desempenho de roteadores de modelos, modelos compostos e outros tipos de lógica de alternância entre múltiplos modelos.

O Auto Router agora usa o Jev Router. Saiba como funciona →
Roteadores
Favicon de nvidia
Favicon de openrouter
Favicon de sakana
Favicon de unbiased
6 roteadores
Auto, Fugu Max, Pareto, Auto Code e Fugu Ultra v2 mais alternância de modelos do Switchyard
Modelos
Favicon de anthropic
Favicon de deepseek
Favicon de google
Favicon de moonshotai
18 modelos
Somente modelos de rota e de referência; as opções de modelo do roteador estão indisponíveis
Resultados
162
execuções publicadas do roteador e da linha de base

Última execução da avaliação comparativa

Fronteira de roteamento de modelos

Por padrão, cada ponto representa a qualidade média de um roteador de modelos nas avaliações comparativas. O custo e a velocidade permanecem no eixo horizontal em vez de serem incorporados à qualidade, ao contrário do Índice de roteamento de modelos na matriz. Selecione uma avaliação comparativa para comparar a qualidade nela ou altere o que o eixo horizontal mede.

Fronteira de qualidade

Custo registrado por tarefa avaliada, em escala logarítmica. A pontuação de qualidade é a acurácia em relação ao melhor roteador em cada avaliação comparativa (0–10), calculada como média entre as avaliações comparativas. Uma referência que supera todos os roteadores tem sua pontuação limitada a 10. Os pontos vazados são linhas de base de modelos individuais, exibidas como referência. As referências com pontuação abaixo do eixo são omitidas; elas continuam na tabela de resultados.

Fronteira de qualidade
Roteador de modelosPontuação de qualidadeCusto / tarefa
Claude Haiku 5.5 (low)3,5US$ 0,053
Claude Haiku 5.5 (medium)4,7US$ 0,098
Switchyard · Claude Opus 5.5 + Claude Sonnet 5.5 (high)8,2US$ 3,02
Claude Opus 5.5 (high)9,1US$ 6,43
Claude Opus 5.5 (low)6,7US$ 0,46
Claude Opus 5.5 (medium)8,1US$ 0,98
Claude Sonnet 5.5 (high)7,5US$ 0,83
Claude Sonnet 5.5 (low)4,8US$ 0,22
Claude Sonnet 5.5 (medium)5,4US$ 0,29
Claude Sonnet 5.58,1US$ 0,92
Switchyard · Deepseek v4.1 Flash + GPT-6-Sol (high)8,4US$ 0,36
Deepseek v4.1 Flash (high)7,9US$ 0,22
Deepseek v4.1 Flash (max)8,2US$ 0,20
Deepseek v4.1 Flash6,7US$ 1,02
Gemini 4 Argon (high)9,3US$ 1,33
Kimi K3 (high)8,9US$ 0,93
Kimi K3 (low)7,4US$ 0,67
Kimi K3.1 Pro (high)9,6US$ 0,50
Kimi K3.1 Pro9,0US$ 2,66
Switchyard · GPT-6-Astra + GPT-6.1-Sol (high)8,6US$ 2,17
GPT-6-Astra (high)9,0US$ 1,73
GPT-6-Astra (low)7,6US$ 1,02
GPT-6-Astra (medium)8,0US$ 1,70
GPT-6-Luna (high)6,5US$ 0,040
GPT-6-Luna (max)7,6US$ 0,074
GPT-6-Luna3,4US$ 0,062
Switchyard · GPT-6-Sol + GPT-6-Astra (high)9,2US$ 1,56
GPT-6-Sol (high)8,5US$ 0,34
Switchyard · GPT-6.1-Sol + GPT-6-Luna (high)7,2US$ 0,37
GPT-6.1-Sol (high)8,2US$ 0,38
GPT-6.1-Sol (low)6,9US$ 0,14
GPT-6.1-Sol (medium)8,4US$ 0,25
GPT-6.1-Terra (high)8,3US$ 0,21
Grok 4.7 (medium)7,0US$ 3,21
Mimo v2.6 Pro (xhigh)7,8US$ 0,15
GLM 5.3 (high)8,0US$ 0,77
Switchyard · GLM 5.3 Flash + Claude Opus 5.5 (high)8,8US$ 1,01
GLM 5.3 Flash (high)6,5US$ 0,15
GLM 5.3 Flash9,3US$ 0,29
Auto (Medium tier)9,3US$ 0,58
Auto Code9,8US$ 1,64
Fugu Max6,8US$ 1,29
Pareto8,6US$ 0,16
Deepseek v4.1 Flash 1017 (high)7,3US$ 0,044
Gemini 3.8 Flash (high)8,6US$ 0,26
Fugu Ultra v28,8US$ 3,95
Claude Haiku 5.5 (high)4,9US$ 0,048
Switchyard · Gemini 4 Argon + Gemini 3.8 Flash (high)8,2US$ 0,70

Matriz de avaliações de desempenho

O índice de roteamento de modelos combina qualidade, velocidade e custo em uma pontuação de 0–10 relativa aos melhores resultados em cada teste de desempenho. Explore a Fronteira de roteamento de modelos.

Roteador de modelos
Roteadores
Auto
Deep SWE69,0%
SWE Atlas QA59,7%
Terminal Bench 4.045,7%
Tau3 BankingMelhor no teste de desempenho53,1%
MMMU Pro Vision80,1%
Auto Code
Deep SWE74,3%
SWE Atlas QA62,5%
Terminal Bench 4.0Melhor no teste de desempenho50,0%
Tau3 Banking52,6%
MMMU Pro Vision82,3%
Fugu Max
Deep SWE33,9%
SWE Atlas QA52,4%
Terminal Bench 4.0—
Tau3 Banking29,9%
MMMU Pro Vision79,5%
Fugu Ultra v2
Deep SWE—
SWE Atlas QAMelhor no teste de desempenho69,5%
Terminal Bench 4.0—
Tau3 Banking35,1%
MMMU Pro Vision80,5%
Pareto
Deep SWE74,3%
SWE Atlas QA60,5%
Terminal Bench 4.042,5%
Tau3 Banking35,1%
MMMU Pro Vision75,5%
Alternância entre múltiplos modelos
Switchyard
Algoritmo: etapa
Claude Opus 5.5 + Claude Sonnet 5.5 (high)
Deep SWE68,8%
SWE Atlas QA56,5%
Terminal Bench 4.045,0%
Tau3 Banking35,1%
MMMU Pro Vision—
Deepseek v4.1 Flash + GPT-6-Sol (high)
Deep SWE65,5%
SWE Atlas QA52,9%
Terminal Bench 4.0—
Tau3 Banking41,2%
MMMU Pro Vision77,0%
Gemini 4 Argon + Gemini 3.8 Flash (high)
Deep SWE—
SWE Atlas QA60,5%
Terminal Bench 4.035,0%
Tau3 Banking47,4%
MMMU Pro Vision—
GLM 5.3 Flash + Claude Opus 5.5 (high)
Deep SWE73,5%
SWE Atlas QA59,7%
Terminal Bench 4.0—
Tau3 Banking38,1%
MMMU Pro Vision79,5%
GPT-6-Astra + GPT-6.1-Sol (high)
Deep SWE73,5%
SWE Atlas QA—
Terminal Bench 4.045,0%
Tau3 Banking38,1%
MMMU Pro Vision—
GPT-6-Sol + GPT-6-Astra (high)
Deep SWE70,8%
SWE Atlas QA66,1%
Terminal Bench 4.0—
Tau3 Banking42,3%
MMMU Pro VisionMelhor no teste de desempenho82,4%
GPT-6.1-Sol + GPT-6-Luna (high)
Deep SWEMelhor no teste de desempenho75,2%
SWE Atlas QA—
Terminal Bench 4.033,3%
Tau3 Banking26,8%
MMMU Pro Vision—
Linhas de base de modelo único (referência)
Claude Haiku 5.5 (high)
Deep SWE—
SWE Atlas QA43,5%
Terminal Bench 4.0—
Tau3 Banking18,6%
MMMU Pro Vision—
Claude Haiku 5.5 (low)
Deep SWE40,7%
SWE Atlas QA24,2%
Terminal Bench 4.0—
Tau3 Banking9,3%
MMMU Pro Vision—
Claude Haiku 5.5 (medium)
Deep SWE52,2%
SWE Atlas QA31,5%
Terminal Bench 4.0—
Tau3 Banking14,4%
MMMU Pro Vision—
Claude Opus 5.5 (high)
Deep SWE70,5%
SWE Atlas QA55,6%
Terminal Bench 4.053,1%
Tau3 Banking42,3%
MMMU Pro Vision81,8%
Claude Opus 5.5 (low)
Deep SWE61,1%
SWE Atlas QA39,5%
Terminal Bench 4.0—
Tau3 Banking33,5%
MMMU Pro Vision—
Claude Opus 5.5 (medium)
Deep SWE67,7%
SWE Atlas QA54,0%
Terminal Bench 4.0—
Tau3 Banking39,7%
MMMU Pro Vision—
Claude Sonnet 5.5
Deep SWE77,8%
SWE Atlas QA—
Terminal Bench 4.031,3%
Tau3 Banking—
MMMU Pro Vision—
Claude Sonnet 5.5 (high)
Deep SWE66,4%
SWE Atlas QA48,4%
Terminal Bench 4.0—
Tau3 Banking35,6%
MMMU Pro Vision—
Claude Sonnet 5.5 (low)
Deep SWE54,0%
SWE Atlas QA20,2%
Terminal Bench 4.0—
Tau3 Banking22,2%
MMMU Pro Vision—
Claude Sonnet 5.5 (medium)
Deep SWE49,6%
SWE Atlas QA33,1%
Terminal Bench 4.0—
Tau3 Banking25,3%
MMMU Pro Vision—
Deepseek v4.1 Flash
Deep SWE63,6%
SWE Atlas QA—
Terminal Bench 4.025,0%
Tau3 Banking—
MMMU Pro Vision—
Deepseek v4.1 Flash (high)
Deep SWE62,8%
SWE Atlas QA63,7%
Terminal Bench 4.0—
Tau3 Banking32,5%
MMMU Pro Vision—
Deepseek v4.1 Flash (max)
Deep SWE61,9%
SWE Atlas QA65,3%
Terminal Bench 4.0—
Tau3 Banking36,1%
MMMU Pro Vision68,2%
Deepseek v4.1 Flash 1017 (high)
Deep SWE—
SWE Atlas QA58,1%
Terminal Bench 4.0—
Tau3 Banking24,7%
MMMU Pro Vision72,7%
Gemini 3.8 Flash (high)
Deep SWE—
SWE Atlas QA43,5%
Terminal Bench 4.0—
Tau3 Banking39,2%
MMMU Pro Vision81,5%
Gemini 4 Argon (high)
Deep SWE76,1%
SWE Atlas QA58,9%
Terminal Bench 4.0—
Tau3 Banking46,4%
MMMU Pro Vision87,0%
GLM 5.3 (high)
Deep SWE64,6%
SWE Atlas QA57,3%
Terminal Bench 4.0—
Tau3 Banking38,1%
MMMU Pro Vision—
GLM 5.3 Flash
Deep SWE84,6%
SWE Atlas QA—
Terminal Bench 4.042,9%
Tau3 Banking—
MMMU Pro Vision—
GLM 5.3 Flash (high)
Deep SWE31,0%
SWE Atlas QA42,7%
Terminal Bench 4.0—
Tau3 Banking34,0%
MMMU Pro Vision75,4%
GPT-6-Astra (high)
Deep SWE73,5%
SWE Atlas QA65,5%
Terminal Bench 4.0—
Tau3 Banking35,6%
MMMU Pro Vision84,2%
GPT-6-Astra (low)
Deep SWE67,2%
SWE Atlas QA54,1%
Terminal Bench 4.0—
Tau3 Banking32,5%
MMMU Pro Vision—
GPT-6-Astra (medium)
Deep SWE69,9%
SWE Atlas QA58,8%
Terminal Bench 4.0—
Tau3 Banking33,0%
MMMU Pro Vision—
GPT-6-Luna
Deep SWE51,2%
SWE Atlas QA—
Terminal Bench 4.00,0%
Tau3 Banking—
MMMU Pro Vision—
GPT-6-Luna (high)
Deep SWE53,1%
SWE Atlas QA35,5%
Terminal Bench 4.0—
Tau3 Banking24,2%
MMMU Pro Vision76,6%
GPT-6-Luna (max)
Deep SWE67,3%
SWE Atlas QA42,7%
Terminal Bench 4.0—
Tau3 Banking30,9%
MMMU Pro Vision77,3%
GPT-6-Sol (high)
Deep SWE61,1%
SWE Atlas QA55,4%
Terminal Bench 4.0—
Tau3 Banking42,3%
MMMU Pro Vision81,5%
GPT-6.1-Sol (high)
Deep SWE70,8%
SWE Atlas QA59,7%
Terminal Bench 4.0—
Tau3 Banking35,1%
MMMU Pro Vision—
GPT-6.1-Sol (low)
Deep SWE60,2%
SWE Atlas QA51,6%
Terminal Bench 4.0—
Tau3 Banking27,8%
MMMU Pro Vision—
GPT-6.1-Sol (medium)
Deep SWE72,6%
SWE Atlas QA60,5%
Terminal Bench 4.0—
Tau3 Banking37,1%
MMMU Pro Vision—
GPT-6.1-Terra (high)
Deep SWE72,8%
SWE Atlas QA53,6%
Terminal Bench 4.040,8%
Tau3 Banking33,0%
MMMU Pro Vision80,3%
Grok 4.7 (medium)
Deep SWE35,4%
SWE Atlas QA58,9%
Terminal Bench 4.0—
Tau3 Banking41,8%
MMMU Pro Vision—
Kimi K3 (high)
Deep SWE69,0%
SWE Atlas QA58,1%
Terminal Bench 4.0—
Tau3 Banking42,8%
MMMU Pro Vision81,0%
Kimi K3 (low)
Deep SWE57,5%
SWE Atlas QA—
Terminal Bench 4.0—
Tau3 Banking38,1%
MMMU Pro Vision—
Kimi K3.1 Pro
Deep SWE60,0%
SWE Atlas QA—
Terminal Bench 4.068,8%
Tau3 Banking—
MMMU Pro Vision—
Kimi K3.1 Pro (high)
Deep SWE77,9%
SWE Atlas QA58,1%
Terminal Bench 4.0—
Tau3 Banking54,6%
MMMU Pro Vision81,8%
Mimo v2.6 Pro (xhigh)
Deep SWE58,4%
SWE Atlas QA54,0%
Terminal Bench 4.0—
Tau3 Banking42,3%
MMMU Pro Vision63,6%
MelhorPiorAs referências de modelos individuais são exibidas para comparação e não são classificadas.

Resultados de roteamento de modelos

Deep SWE43 resultados
Deep SWE GLM 5.3 Flash84,6%US$ 0,20US$ 0,2426,7m
Deep SWE Kimi K3.1 Pro77,9%US$ 0,85US$ 1,0918,6m
Deep SWE Claude Sonnet 5.577,8%US$ 0,40US$ 0,516,0m
Deep SWE Gemini 4 Argon76,1%US$ 3,12US$ 4,1026,6m
Deep SWE Switchyard · GPT-6.1-Sol + GPT-6-Luna (high)75,2%US$ 0,47US$ 0,6312,6m
Deep SWE Auto Code74,3%US$ 1,98US$ 2,6615,2m
Deep SWE Pareto74,3%US$ 0,19US$ 0,267,5m
Deep SWE Switchyard · GPT-6-Astra + GPT-6.1-Sol (high)73,5%US$ 2,54US$ 3,4612,5m
Deep SWE GPT-6-Astra73,5%US$ 2,99US$ 4,0711,7m
Deep SWE Switchyard · GLM 5.3 Flash + Claude Opus 5.5 (high)73,5%US$ 2,47US$ 3,3617,4m
Deep SWE GPT-6.1-Terra72,8%US$ 0,31US$ 0,438,3m
Deep SWE GPT-6.1-Sol · openai72,6%US$ 0,30US$ 0,413,7m
Deep SWE Switchyard · GPT-6-Sol + GPT-6-Astra (high)70,8%US$ 2,35US$ 3,328,5m
Deep SWE GPT-6.1-Sol (high)70,8%US$ 0,46US$ 0,656,5m
Deep SWE Claude Opus 5.570,5%US$ 7,76US$ 11,0051,9m
Deep SWE GPT-6-Astra69,9%US$ 1,99US$ 2,855,8m
Deep SWE Kimi K3 (high)69,0%US$ 1,85US$ 2,6823,1m
Deep SWE Auto (Medium tier)69,0%US$ 0,57US$ 0,839,6m
Deep SWE Switchyard · Claude Opus 5.5 + Claude Sonnet 5.5 (high)68,8%US$ 6,59US$ 9,5944,9m
Deep SWE Claude Opus 5.567,7%US$ 1,54US$ 2,277,0m
Deep SWE GPT-6-Luna67,3%US$ 0,18US$ 0,2717,7m
Deep SWE GPT-6-Astra67,2%US$ 1,02US$ 1,523,1m
Deep SWE Claude Sonnet 5.566,4%US$ 1,86US$ 2,8021,4m
Deep SWE Switchyard · Deepseek v4.1 Flash + GPT-6-Sol (high)65,5%US$ 0,48US$ 0,746,3m
Deep SWE GLM 5.3 (high)64,6%US$ 1,46US$ 2,2616,3m
Deep SWE Deepseek v4.1 Flash63,6%US$ 0,21US$ 0,3315,6m
Deep SWE Deepseek v4.1 Flash62,8%US$ 0,37US$ 0,6019,9m
Deep SWE Deepseek v4.1 Flash61,9%US$ 0,42US$ 0,6827,5m
Deep SWE Claude Opus 5.561,1%US$ 0,52US$ 0,852,6m
Deep SWE GPT-6-Sol61,1%US$ 0,46US$ 0,753,6m
Deep SWE GPT-6.1-Sol60,2%US$ 0,14US$ 0,233,7m
Deep SWE Kimi K3.1 Pro60,0%US$ 1,43US$ 2,3824,9m
Deep SWE Mimo v2.6 Pro58,4%US$ 0,31US$ 0,5343,2m
Deep SWE Kimi K357,5%US$ 0,63US$ 1,104,4m
Deep SWE Claude Sonnet 5.554,0%US$ 0,36US$ 0,664,9m
Deep SWE GPT-6-Luna53,1%US$ 0,11US$ 0,2115,8m
Deep SWE Claude Haiku 5.552,2%US$ 0,25US$ 0,4823,7m
Deep SWE GPT-6-Luna51,2%US$ 0,091US$ 0,1810,8m
Deep SWE Claude Sonnet 5.549,6%US$ 0,51US$ 1,037,0m
Deep SWE Claude Haiku 5.540,7%US$ 0,14US$ 0,3416,2m
Deep SWE Grok 4.735,4%US$ 5,89US$ 16,6412,2m
Deep SWE Fugu Max33,9%US$ 3,00US$ 8,8525,4m
Deep SWE GLM 5.3 Flash (high)31,0%US$ 0,34US$ 1,1027,0m
MMMU Pro Vision22 resultados
MMMU Pro Vision Gemini 4 Argon87,0%US$ 0,027US$ 0,03136s
MMMU Pro Vision GPT-6-Astra84,2%US$ 0,058US$ 0,0698s
MMMU Pro Vision Switchyard · GPT-6-Sol + GPT-6-Astra (high)82,4%US$ 0,041US$ 0,0499s
MMMU Pro Vision Auto Code82,3%US$ 0,025US$ 0,0316s
MMMU Pro Vision Claude Opus 5.581,8%US$ 0,026US$ 0,0329s
MMMU Pro Vision Kimi K3.1 Pro81,8%US$ 0,016US$ 0,01929s
MMMU Pro Vision Gemini 3.8 Flash81,5%US$ 0,023US$ 0,02945s
MMMU Pro Vision GPT-6-Sol81,5%US$ 0,013US$ 0,0167s
MMMU Pro Vision Kimi K381,0%US$ 0,021US$ 0,02588s
MMMU Pro Vision Fugu Ultra v280,5%US$ 0,11US$ 0,1442s
MMMU Pro Vision GPT-6.1-Terra80,3%US$ 0,006US$ 0,0076s
MMMU Pro Vision Auto (Medium tier)80,1%US$ 0,020US$ 0,02437s
MMMU Pro Vision Switchyard · GLM 5.3 Flash + Claude Opus 5.5 (high)79,5%US$ 0,017US$ 0,02123s
MMMU Pro Vision Fugu Max79,5%US$ 0,019US$ 0,02424s
MMMU Pro Vision GPT-6-Luna77,3%US$ 0,002US$ 0,00338s
MMMU Pro Vision Switchyard · Deepseek v4.1 Flash + GPT-6-Sol (high)77,0%US$ 0,011US$ 0,01442s
MMMU Pro Vision GPT-6-Luna76,6%US$ 0,00089US$ 0,0018s
MMMU Pro Vision Pareto75,5%US$ 0,015US$ 0,01917s
MMMU Pro Vision GLM 5.3 Flash75,4%US$ 0,0008US$ 0,00133s
MMMU Pro Vision Deepseek v4.1 Flash 101772,7%US$ 0,006US$ 0,00949s
MMMU Pro Vision Deepseek v4.1 Flash68,2%US$ 0,016US$ 0,0232,1m
MMMU Pro Vision Mimo v2.6 Pro63,6%US$ 0,011US$ 0,0172,5m
SWE Atlas QA40 resultados
SWE Atlas QA Fugu Ultra v269,5%US$ 8,91US$ 12,8215,5m
SWE Atlas QA Switchyard · GPT-6-Sol + GPT-6-Astra (high)66,1%US$ 1,98US$ 2,995,3m
SWE Atlas QA GPT-6-Astra (high)65,5%US$ 2,06US$ 3,144,8m
SWE Atlas QA Deepseek v4.1 Flash65,3%US$ 0,16US$ 0,2512,7m
SWE Atlas QA Deepseek v4.1 Flash63,7%US$ 0,19US$ 0,2911,0m
SWE Atlas QA Auto Code62,5%US$ 0,66US$ 1,054,3m
SWE Atlas QA Switchyard · Gemini 4 Argon + Gemini 3.8 Flash (high)60,5%US$ 1,28US$ 2,1112,6m
SWE Atlas QA GPT-6.1-Sol60,5%US$ 0,22US$ 0,363,7m
SWE Atlas QA Pareto60,5%US$ 0,14US$ 0,234,8m
SWE Atlas QA GPT-6.1-Sol59,7%US$ 0,39US$ 0,655,8m
SWE Atlas QA Switchyard · GLM 5.3 Flash + Claude Opus 5.5 (high)59,7%US$ 0,87US$ 1,467,1m
SWE Atlas QA Auto (Medium tier)59,7%US$ 0,36US$ 0,604,9m
SWE Atlas QA Gemini 4 Argon58,9%US$ 1,25US$ 2,1312,9m
SWE Atlas QA Grok 4.758,9%US$ 1,44US$ 2,445,3m
SWE Atlas QA GPT-6-Astra58,8%US$ 1,63US$ 2,774,1m
SWE Atlas QA Deepseek v4.1 Flash 101758,1%US$ 0,095US$ 0,165,7m
SWE Atlas QA Kimi K3 (high)58,1%US$ 0,86US$ 1,498,1m
SWE Atlas QA Kimi K3.1 Pro58,1%US$ 0,47US$ 0,809,5m
SWE Atlas QA GLM 5.3 (high)57,3%US$ 0,64US$ 1,139,1m
SWE Atlas QA Switchyard · Claude Opus 5.5 + Claude Sonnet 5.5 (high)56,5%US$ 0,80US$ 1,427,0m
SWE Atlas QA Claude Opus 5.5 (high)55,6%US$ 0,79US$ 1,433,8m
SWE Atlas QA GPT-6-Sol (high)55,4%US$ 0,35US$ 0,642,6m
SWE Atlas QA GPT-6-Astra54,1%US$ 0,89US$ 1,642,6m
SWE Atlas QA Claude Opus 5.554,0%US$ 0,48US$ 0,892,8m
SWE Atlas QA Mimo v2.6 Pro54,0%US$ 0,15US$ 0,2820,4m
SWE Atlas QA GPT-6.1-Terra53,6%US$ 0,16US$ 0,293,6m
SWE Atlas QA Switchyard · Deepseek v4.1 Flash + GPT-6-Sol (high)52,9%US$ 0,36US$ 0,683,7m
SWE Atlas QA Fugu Max52,4%US$ 1,56US$ 2,9811,1m
SWE Atlas QA GPT-6.1-Sol51,6%US$ 0,13US$ 0,252,6m
SWE Atlas QA Claude Sonnet 5.548,4%US$ 0,26US$ 0,543,2m
SWE Atlas QA Claude Haiku 5.543,5%US$ 0,048US$ 0,115,7m
SWE Atlas QA Gemini 3.8 Flash

Falhou

43,5%US$ 1,04US$ 2,409,8m
SWE Atlas QA GPT-6-Luna42,7%US$ 0,064US$ 0,157,8m
SWE Atlas QA GLM 5.3 Flash (high)42,7%US$ 0,20US$ 0,4615,1m
SWE Atlas QA Claude Opus 5.539,5%US$ 0,16US$ 0,4167s
SWE Atlas QA GPT-6-Luna (high)35,5%US$ 0,027US$ 0,0772,8m
SWE Atlas QA Claude Sonnet 5.533,1%US$ 0,098US$ 0,3089s
SWE Atlas QA Claude Haiku 5.531,5%US$ 0,015US$ 0,0482,5m
SWE Atlas QA Claude Haiku 5.524,2%US$ 0,007US$ 0,03087s
SWE Atlas QA Claude Sonnet 5.520,2%US$ 0,069US$ 0,3468s
Tau3 Banking43 resultados
Tau3 Banking Kimi K3.1 Pro54,6%US$ 0,66US$ 1,216,0m
Tau3 Banking Auto (Medium tier)53,1%US$ 0,60US$ 1,135,4m
Tau3 Banking Auto Code52,6%US$ 0,60US$ 1,135,9m
Tau3 Banking Switchyard · Gemini 4 Argon + Gemini 3.8 Flash (high)47,4%US$ 0,74US$ 1,577,0m
Tau3 Banking Gemini 4 Argon46,4%US$ 0,94US$ 2,0310,5m
Tau3 Banking Kimi K342,8%US$ 0,97US$ 2,273,4m
Tau3 Banking Claude Opus 5.542,3%US$ 0,98US$ 2,321,9m
Tau3 Banking Switchyard · GPT-6-Sol + GPT-6-Astra (high)42,3%US$ 1,86US$ 4,412,6m
Tau3 Banking GPT-6-Sol (high)42,3%US$ 0,55US$ 1,291,6m
Tau3 Banking Mimo v2.6 Pro42,3%US$ 0,12US$ 0,2911,3m
Tau3 Banking Grok 4.741,8%US$ 2,32US$ 5,554,5m
Tau3 Banking Switchyard · Deepseek v4.1 Flash + GPT-6-Sol (high)41,2%US$ 0,59US$ 1,432,8m
Tau3 Banking Claude Opus 5.539,7%US$ 0,92US$ 2,321,8m
Tau3 Banking Gemini 3.8 Flash39,2%US$ 0,50US$ 1,274,2m
Tau3 Banking Kimi K338,1%US$ 0,71US$ 1,862,6m
Tau3 Banking Switchyard · GPT-6-Astra + GPT-6.1-Sol (high)38,1%US$ 1,37US$ 3,602,4m
Tau3 Banking GLM 5.3 (high)38,1%US$ 0,19US$ 0,491,9m
Tau3 Banking Switchyard · GLM 5.3 Flash + Claude Opus 5.5 (high)38,1%US$ 0,67US$ 1,743,8m
Tau3 Banking GPT-6.1-Sol37,1%US$ 0,23US$ 0,6267s
Tau3 Banking Deepseek v4.1 Flash36,1%US$ 0,20US$ 0,5617,9m
Tau3 Banking Claude Sonnet 5.535,6%US$ 0,36US$ 1,0165s
Tau3 Banking GPT-6-Astra35,6%US$ 1,83US$ 5,142,1m
Tau3 Banking Switchyard · Claude Opus 5.5 + Claude Sonnet 5.5 (high)35,1%US$ 0,78US$ 2,221,7m
Tau3 Banking GPT-6.1-Sol35,1%US$ 0,30US$ 0,861,5m
Tau3 Banking Fugu Ultra v235,1%US$ 2,84US$ 8,114,1m
Tau3 Banking Pareto35,1%US$ 0,15US$ 0,441,7m
Tau3 Banking GLM 5.3 Flash34,0%US$ 0,043US$ 0,1378s
Tau3 Banking Claude Opus 5.533,5%US$ 0,69US$ 2,0785s
Tau3 Banking GPT-6-Astra33,0%US$ 1,49US$ 4,5188s
Tau3 Banking GPT-6.1-Terra33,0%US$ 0,17US$ 0,5376s
Tau3 Banking Deepseek v4.1 Flash32,5%US$ 0,11US$ 0,337,5m
Tau3 Banking GPT-6-Astra32,5%US$ 1,16US$ 3,5881s
Tau3 Banking GPT-6-Luna30,9%US$ 0,046US$ 0,153,8m
Tau3 Banking Fugu Max29,9%US$ 0,59US$ 1,983,3m
Tau3 Banking GPT-6.1-Sol27,8%US$ 0,16US$ 0,5857s
Tau3 Banking Switchyard · GPT-6.1-Sol + GPT-6-Luna (high)26,8%US$ 0,21US$ 0,782,6m
Tau3 Banking Claude Sonnet 5.525,3%US$ 0,27US$ 1,0852s
Tau3 Banking Deepseek v4.1 Flash 101724,7%US$ 0,032US$ 0,132,2m
Tau3 Banking GPT-6-Luna24,2%US$ 0,020US$ 0,0831,5m
Tau3 Banking Claude Sonnet 5.522,2%US$ 0,24US$ 1,0847s
Tau3 Banking Claude Haiku 5.518,6%US$ 0,048US$ 0,2678s
Tau3 Banking Claude Haiku 5.514,4%US$ 0,025US$ 0,1859s
Tau3 Banking Claude Haiku 5.59,3%US$ 0,016US$ 0,1747s
Terminal Bench 4.014 resultados
Terminal Bench 4.0 Kimi K3.1 Pro68,8%US$ 3,90US$ 5,6773,3m
Terminal Bench 4.0 Claude Opus 5.553,1%US$ 22,60US$ 35,72198,6m
Terminal Bench 4.0 Auto Code50,0%US$ 4,96US$ 6,6570,8m
Terminal Bench 4.0 Auto (Medium tier)45,7%US$ 1,33US$ 2,4025,1m
Terminal Bench 4.0 Switchyard · Claude Opus 5.5 + Claude Sonnet 5.5 (high)45,0%US$ 3,89US$ 8,6550,2m
Terminal Bench 4.0 Switchyard · GPT-6-Astra + GPT-6.1-Sol (high)45,0%US$ 2,60US$ 5,7813,0m
Terminal Bench 4.0 GLM 5.3 Flash42,9%US$ 0,39US$ 0,9185,9m
Terminal Bench 4.0 Pareto42,5%US$ 0,29US$ 0,6922,8m
Terminal Bench 4.0 GPT-6.1-Terra40,8%US$ 0,40US$ 0,9823,8m
Terminal Bench 4.0 Switchyard · Gemini 4 Argon + Gemini 3.8 Flash (high)35,0%US$ 0,083US$ 0,2457,4m
Terminal Bench 4.0 Switchyard · GPT-6.1-Sol + GPT-6-Luna (high)33,3%US$ 0,44US$ 1,3311,9m
Terminal Bench 4.0 Claude Sonnet 5.531,3%US$ 1,44US$ 4,6159,2m
Terminal Bench 4.0 Deepseek v4.1 Flash25,0%US$ 1,84US$ 7,34148,3m
Terminal Bench 4.0 GPT-6-Luna0,0%US$ 0,033--30,2m