Ranking · agentes
¿Qué modelo usar para tus agentes?
Cruzamos los votos reales de Arena en tareas de agente con los benchmarks, el precio y la velocidad de Artificial Analysis. Leaderboard del 16 de octubre de 2026.
Puntuación global de Arena en tareas de agente (votos reales)
| # | Modelo | Arena agente | Intelligence Index | $/M tokens |
|---|---|---|---|---|
| 1 | Claude Opus 5.5 (Max)Pareto anthropicProprietary | +18,9 | 57,6 | $8 |
| 2 | GPT-6 Astra (Max) openaiProprietary | +14,1 | 52,7 | $20 |
| 3 | Claude Fable 5.1 (Max) anthropicProprietary | +14,0 | 53,4 | $20 |
| 4 | GPT-6.1 Sol (Max)Pareto openaiProprietary | +13,6 | 51,8 | $4 |
| 5 | Gemini 4 Argon (High) googleProprietary | +13,2 | 52,6 | $4 |
| 6 | Claude Sonnet 5.5 (Max) anthropicProprietary | +13,0 | 56,0 | $4 |
| 7 | GPT-6 Sol (Max) openaiProprietary | +10,3 | 47,6 | $4 |
| 8 | Claude Opus 5 (Max) anthropicProprietary | +8,7 | 50,8 | $10 |
| 9 | Claude Fable 5 (High) anthropicProprietary | +8,2 | · | · |
| 10 | GPT-5.6 Sol (xHigh) openaiProprietary | +7,1 | 44,0 | $8 |
| 11 | Muse Wonder (Max) metaProprietary | +4,9 | 52,4 | $8 |
| 12 | Claude Sonnet 5 (High) anthropicProprietary | +4,6 | 31,7 | $4 |
| 13 | MiMo-V2.6-ProPareto xiaomiMIT | +4,0 | 46,3 | $0,54 |
| 14 | Muse Spark 1.3 (Max) metaProprietary | +3,8 | 48,1 | $2 |
| 15 | Gemini 3.8 Flash (High) googleProprietary | +3,7 | 40,9 | $1,5 |
| 16 | Grok 4.7 (xHigh) xaiProprietary | +3,2 | 46,4 | $3 |
| 17 | GPT-5.5 (xHigh) openaiProprietary | +3,1 | 38,4 | $11,25 |
| 18 | GPT-6 Luna (Max)Pareto openaiProprietary | +3,0 | 38,1 | $0,2 |
| 19 | DeepSeek-V4.1-Flash (Max) deepseekMIT | +2,7 | 39,5 | $0,53 |
| 20 | Hy4 preview tencentApache 2.0 | +2,7 | · | · |
| 21 | MiMo-V2.6-FlashPareto xiaomiMIT | +2,6 | 37,9 | $0,18 |
| 22 | GLM-5.2 (Max) zaiMIT | +2,3 | 33,7 | $2,15 |
| 23 | Kimi K3 (Max) moonshotKimi K3 license | +1,9 | 43,6 | $6 |
| 24 | GLM-5.3 (Max) zaiMIT | +1,8 | 44,8 | $2,15 |
| 25 | Qwen3.8-Max alibabaProprietary | +1,8 | 40,2 | $3 |
| 26 | Claude Haiku 5.5 (High) anthropicProprietary | +1,5 | 37,8 | $0,2 |
| 27 | GPT-5.4 (High) openaiProprietary | +1,4 | · | · |
| 28 | Grok 4.6 (xHigh) xaiProprietary | +0,9 | 44,2 | $3 |
| 29 | GPT-5.6 Terra (xHigh) openaiProprietary | +0,4 | 38,0 | $4,5 |
| 30 | DeepSeek-V4-Pro (High, 20260813) deepseekMIT | +0,2 | · | · |
| 31 | GPT-5.5 openaiProprietary | -0,2 | · | · |
| 32 | Gemini 3.7 Flash (High) googleProprietary | -0,7 | 39,1 | $1,5 |
| 33 | GLM-5.3-Flash zaiMIT | -1,2 | 41,8 | $0,24 |
| 34 | Step 5 Preview stepfunProprietary | -1,4 | 43,7 | $1,43 |
| 35 | GPT-5.6 Luna (xHigh) openaiProprietary | -2,0 | 34,6 | $0,45 |
| 36 | Qwen3.8-Flash-Next alibabaqwen-community-1.0 | -2,9 | 39,8 | $0,23 |
| 37 | Qwen3.8-27B alibabaApache 2.0 | -3,4 | · | · |
| 38 | Beam 1.0 501B A23B reflectionApache 2.0 | -3,7 | 30,3 | $1,05 |
| 39 | Gemini 3.1 Pro Preview googleProprietary | -4,0 | 29,7 | $4,5 |
| 40 | Mistral Large 4 mistralProprietary | -4,0 | · | · |
| 41 | Hy3 tencentApache 2.0 | -6,0 | 25,3 | $0,24 |
| 42 | Qwen3.7-Max alibabaProprietary | -6,3 | 29,5 | $3,75 |
| 43 | Muse Spark 1.2 (xHigh) metaProprietary | -6,4 | 39,6 | $2 |
| 44 | Muse Spark 1.1 metaProprietary | -6,4 | 33,7 | $2 |
| 45 | Qwen3.7-Plus alibabaProprietary | -6,7 | 25,2 | $0,7 |
| 46 | MiMo-V2.5-Pro xiaomiMIT | -8,0 | 26,0 | $0,54 |
| 47 | MiniMax-M3 minimaxMiniMax Community License | -8,3 | 29,2 | $0,53 |
| 48 | Nemotron 3 Ultra nvidiaOpenMDW-1.1 | -8,9 | 22,9 | $1,13 |
| 49 | Mistral Medium 3.5 mistralModified MIT | -11,7 | 14,2 | $3 |
| 50 | Inkling thinkyApache 2.0 | -14,4 | 25,0 | $1,76 |
| 51 | Inkling-Small thinkyApache 2.0 | -15,1 | 25,7 | $0,53 |
| 52 | Solar Pro 4 upstageProprietary | -20,1 | 26,6 | $0,53 |
Calidad frente a precio
La línea verde une los modelos que nadie supera a la vez en Arena agente y en precio: si un modelo queda por debajo, existe otro igual de barato que rinde más. Azul: pesos abiertos. Rojo: propietarios.