Anuncio

SRExpert

Deja de saltar entre 6 herramientas de Kubernetes.

Monitoreo, alertas, seguridad, despliegues Helm y operaciones con IA en una sola plataforma.

Probar 30 días gratis

Comparación de costos

¿API o GPU propia?

En los modelos con open weights, la respuesta depende del volumen. Por debajo de cierto throughput gana la API: no pagas nada mientras está inactiva. Por encima, el costo fijo de la tarjeta alquilada se paga solo.

167modelos

Recolectado 12 nov

El punto de equilibrio es aritmética: el costo por hora de la tarjeta más barata que soporta el modelo, dividido por lo que cuestan los mismos tokens en la API. La mezcla supone 3 tokens de entrada por cada 1 de salida, lo típico en un chat.

Lo que esto no te dice es si tu configuración alcanza ese throughput. Los tokens por hora dependen de la cuantización, el contexto y el batching: eso es medición, y no lo hemos medido.

167 de 167

ModeloAPIAlquilar gana por encima de
Nex AGI: Nex-N2.6nex-agi0,19 US$/M tokens8M tok/h
LiquidAI: LFM3-1.7Bliquid0,11 US$/M tokens0.24M tok/h
Ricursive: Ripple 65B Previewricursiveai0,30 US$/M tokens8M tok/h
Motif Technologies: Motif 3motif-technologies0,525 US$/M tokens3M tok/h
MoonshotAI: Kimi K3.1 Promoonshotai4,00 US$/M tokens—
MiniMax: MiniMax M3.1 Flashminimax0,131 US$/M tokens7M tok/h
DeepSeek: DeepSeek V4.1 Flash 1017 (batch)deepseek0,058 US$/M tokens565M tok/h
DeepSeek: DeepSeek V4.1 Flash 1017deepseek0,13 US$/M tokens254M tok/h
Reflection: Beam 1.0 501B A23Breflection0,475 US$/M tokens30M tok/h
Meta: Muse Spark Openmeta2,00 US$/M tokens16M tok/h
Moondream: Moondream 3 Previewmoondream0,85 US$/M tokens0.08M tok/h
Moondream: Moondream 3.1moondream0,85 US$/M tokens0.08M tok/h
NVIDIA: Nemotron 3.5 Super VLnvidia0,23 US$/M tokens24M tok/h
LiquidAI: LFM2.5-24B-A2Bliquid0,213 US$/M tokens1M tok/h
StepFun: Step 5 Previewstepfun1,425 US$/M tokens12M tok/h
inclusionAI: Ling 3.1 Flashinclusionai0,068 US$/M tokens248M tok/h
Cohere: Command A+cohere0,60 US$/M tokens2M tok/h
Xiaomi: MiMo-V2.6-Flashxiaomi0,175 US$/M tokens4M tok/h
Xiaomi: MiMo-V2.6-Proxiaomi0,544 US$/M tokens26M tok/h
DeepSeek: DeepSeek V4.1 Flashdeepseek0,18 US$/M tokens183M tok/h
inclusionAI: Ling 3.0 Flash VLinclusionai0,09 US$/M tokens61M tok/h
DeepSeek: DeepSeek V4.1 Flash (batch)deepseek0,168 US$/M tokens196M tok/h
Nex AGI: Nex-N2.5-Mininex-agi0,044 US$/M tokens5M tok/h
Nex AGI: Nex-N2.5-Pronex-agi0,119 US$/M tokens16M tok/h
IBM: Granite 4.2 8Bibm-granite0,108 US$/M tokens0.62M tok/h
Tencent: Hy4 previewtencent1,126 US$/M tokens29M tok/h
Z.ai: GLM 5.3 Flash (batch)z-ai0,051 US$/M tokens29M tok/h
Qwen: Qwen3.8 Flashqwen0,23 US$/M tokens4M tok/h
Z.ai: GLM 5.3 Flashz-ai0,05 US$/M tokens29M tok/h
DeepSeek: DeepSeek V4 Flash Vision Expdeepseek0,66 US$/M tokens2M tok/h
Tencent: Hy-MT2-30B-A3Btencent0,129 US$/M tokens2M tok/h
Tencent: Hy-MT2-1.8Btencent0,077 US$/M tokens0.35M tok/h
Tencent: Hy-MT2-7Btencent0,129 US$/M tokens0.52M tok/h
Z.ai: GLM 5.3z-ai0,402 US$/M tokens82M tok/h
Z.ai: GLM 5.3 (batch)z-ai0,838 US$/M tokens39M tok/h
Qwen: Qwen3.8 27Bqwen0,581 US$/M tokens0.41M tok/h
Qwen: Qwen3.8 27B (batch)qwen0,62 US$/M tokens0.39M tok/h
Qwen: Qwen3.8 2.4T A95Bqwen3,00 US$/M tokens—
DeepSeek: DeepSeek V4 Pro 0813deepseek1,193 US$/M tokens—
NVIDIA: Nemotron 3.5 Lightningnvidia0,103 US$/M tokens2M tok/h
LiquidAI: LFM2.5-2.6Bliquid0,387 US$/M tokens0.07M tok/h
Meta: Muse Glimmer 30Bmeta0,525 US$/M tokens0.46M tok/h
Meta: Muse Glimmer 30B (batch)meta0,42 US$/M tokens0.57M tok/h
inclusionAI: Ling 3.0 Tinyinclusionai0,023 US$/M tokens3M tok/h
DeepSeek: DeepSeek V4 Flash 0731 (batch)deepseek0,072 US$/M tokens20M tok/h
DeepSeek: DeepSeek V4 Flash 0731deepseek0,108 US$/M tokens13M tok/h
Thinking Machines: Inkling Small (batch)thinkingmachines0,51 US$/M tokens2M tok/h
Thinking Machines: Inkling Smallthinkingmachines0,638 US$/M tokens2M tok/h
inclusionAI: Ling 3.0 Flashinclusionai0,032 US$/M tokens174M tok/h
Meituan: LongCat 2.0meituan0,525 US$/M tokens—
Thinking Machines: Inklingthinkingmachines1,763 US$/M tokens22M tok/h
MoonshotAI: Kimi K3moonshotai4,144 US$/M tokens—
MoonshotAI: Kimi K3 (batch)moonshotai4,56 US$/M tokens—
Tencent: Hy3tencent0,144 US$/M tokens10M tok/h
Z.ai: GLM 5.2z-ai1,705 US$/M tokens19M tok/h
Z.ai: GLM 5.2 (batch)z-ai0,664 US$/M tokens50M tok/h
MoonshotAI: Kimi K2.7 Codemoonshotai1,325 US$/M tokens33M tok/h
NVIDIA: Nemotron 3.5 Content Safetynvidia0,20 US$/M tokens0.13M tok/h
NVIDIA: Nemotron 3 Ultranvidia1,05 US$/M tokens16M tok/h
MiniMax: MiniMax M3minimax0,525 US$/M tokens4M tok/h
MiniMax: MiniMax M3 (batch)minimax0,388 US$/M tokens5M tok/h
StepFun: Step 3.7 Flashstepfun0,438 US$/M tokens2M tok/h
Mistral: Mistral Medium 3.5mistralai3,00 US$/M tokens2M tok/h
Mistral: Mistral Medium 3.5 (batch)mistralai1,50 US$/M tokens4M tok/h
Qwen: Qwen3.6 35B A3Bqwen0,30 US$/M tokens0.80M tok/h
Qwen: Qwen3.6 27Bqwen1,04 US$/M tokens0.23M tok/h
DeepSeek: DeepSeek V4 Pro 0423deepseek1,194 US$/M tokens—
DeepSeek: DeepSeek V4 Flash 0423deepseek0,113 US$/M tokens13M tok/h
Tencent: Hy3 previewtencent0,285 US$/M tokens5M tok/h
Xiaomi: MiMo-V2.5-Proxiaomi0,81 US$/M tokens54M tok/h
Xiaomi: MiMo-V2.5xiaomi0,80 US$/M tokens2M tok/h
MoonshotAI: Kimi K2.6moonshotai1,34 US$/M tokens33M tok/h
Z.ai: GLM 5.1z-ai2,043 US$/M tokens16M tok/h
Google: Gemma 4 26B A4B google0,085 US$/M tokens3M tok/h
Google: Gemma 4 31Bgoogle0,25 US$/M tokens0.96M tok/h
Reka Edgerekaai0,10 US$/M tokens0.45M tok/h
MiniMax: MiniMax M2.7minimax0,368 US$/M tokens3M tok/h
Mistral: Mistral Small 4mistralai0,262 US$/M tokens21M tok/h
NVIDIA: Nemotron 3 Supernvidia0,156 US$/M tokens35M tok/h
Qwen: Qwen3.5-9Bqwen0,19 US$/M tokens0.35M tok/h
Qwen: Qwen3.5-35B-A3Bqwen0,355 US$/M tokens0.68M tok/h
Qwen: Qwen3.5-27Bqwen0,845 US$/M tokens0.28M tok/h
Qwen: Qwen3.5-122B-A10Bqwen0,715 US$/M tokens8M tok/h
Qwen: Qwen3.5 397B A17Bqwen1,35 US$/M tokens1M tok/h
MiniMax: MiniMax M2.5minimax0,473 US$/M tokens3M tok/h
Z.ai: GLM 5z-ai1,55 US$/M tokens21M tok/h
Qwen: Qwen3 Coder Nextqwen0,29 US$/M tokens8M tok/h
StepFun: Step 3.5 Flashstepfun0,15 US$/M tokens6M tok/h
MoonshotAI: Kimi K2.5moonshotai1,23 US$/M tokens36M tok/h
Z.ai: GLM 4.7 Flashz-ai0,145 US$/M tokens2M tok/h
Z.ai: GLM 4.7z-ai1,00 US$/M tokens2M tok/h
NVIDIA: Nemotron 3 Nano 30B A3Bnvidia0,105 US$/M tokens2M tok/h
Mistral: Devstral 2 2512mistralai0,80 US$/M tokens7M tok/h
Z.ai: GLM 4.6Vz-ai0,45 US$/M tokens12M tok/h
Mistral: Ministral 3 14B 2512mistralai0,20 US$/M tokens1M tok/h
Mistral: Ministral 3 8B 2512mistralai0,15 US$/M tokens0.45M tok/h
Mistral: Ministral 3 3B 2512mistralai0,10 US$/M tokens0.27M tok/h
DeepSeek: DeepSeek V3.2deepseek0,29 US$/M tokens66M tok/h
MoonshotAI: Kimi K2 Thinkingmoonshotai1,075 US$/M tokens41M tok/h
Mistral: Voxtral Small 24B 2507mistralai0,15 US$/M tokens2M tok/h
OpenAI: gpt-oss-safeguard-20bopenai0,131 US$/M tokens2M tok/h
Qwen: Qwen3 VL 8B Instructqwen0,375 US$/M tokens0.18M tok/h
Qwen: Qwen3 VL 30B A3B Thinkingqwen0,467 US$/M tokens0.51M tok/h
Qwen: Qwen3 VL 30B A3B Instructqwen0,262 US$/M tokens0.92M tok/h
Z.ai: GLM 4.6z-ai0,875 US$/M tokens2M tok/h
DeepSeek: DeepSeek V3.2 Expdeepseek0,305 US$/M tokens63M tok/h
TheDrummer: Cydonia 24B V4.1thedrummer0,35 US$/M tokens0.69M tok/h
Qwen: Qwen3 VL 235B A22B Instructqwen0,633 US$/M tokens2M tok/h
DeepSeek: DeepSeek V3.1 Terminusdeepseek0,453 US$/M tokens42M tok/h
Qwen: Qwen3 Next 80B A3B Instructqwen0,343 US$/M tokens7M tok/h
MoonshotAI: Kimi K2 0905moonshotai1,075 US$/M tokens41M tok/h
Nous: Hermes 4 405Bnousresearch1,50 US$/M tokens1M tok/h
DeepSeek: DeepSeek V3.1deepseek0,425 US$/M tokens45M tok/h
Z.ai: GLM 4.5Vz-ai0,90 US$/M tokens6M tok/h
OpenAI: gpt-oss-120bopenai0,262 US$/M tokens21M tok/h
OpenAI: gpt-oss-120b (batch)openai0,056 US$/M tokens98M tok/h
OpenAI: gpt-oss-20bopenai0,055 US$/M tokens4M tok/h
OpenAI: gpt-oss-20b (batch)openai0,046 US$/M tokens5M tok/h
Qwen: Qwen3 Coder 30B A3B Instructqwen0,262 US$/M tokens0.92M tok/h
Qwen: Qwen3 30B A3B Instruct 2507qwen0,15 US$/M tokens2M tok/h
Z.ai: GLM 4.5z-ai1,00 US$/M tokens2M tok/h
Z.ai: GLM 4.5 Airz-ai0,425 US$/M tokens13M tok/h
Qwen: Qwen3 235B A22B Thinking 2507qwen1,213 US$/M tokens0.99M tok/h
Qwen: Qwen3 Coder 480B A35Bqwen0,475 US$/M tokens30M tok/h
ByteDance: UI-TARS 7B bytedance0,125 US$/M tokens0.54M tok/h
Qwen: Qwen3 235B A22B Instruct 2507qwen0,305 US$/M tokens4M tok/h
MoonshotAI: Kimi K2 0711moonshotai1,003 US$/M tokens44M tok/h
Venice: Uncensoredcognitivecomputations0,375 US$/M tokens0.64M tok/h
Tencent: Hunyuan A13B Instructtencent0,248 US$/M tokens10M tok/h
Mistral: Mistral Small 3.2 24Bmistralai0,15 US$/M tokens2M tok/h
DeepSeek: R1 0528deepseek0,913 US$/M tokens21M tok/h
Meta: Llama Guard 4 12Bmeta-llama0,18 US$/M tokens1M tok/h
Qwen: Qwen3 30B A3Bqwen0,215 US$/M tokens1M tok/h
Qwen: Qwen3 14Bqwen0,13 US$/M tokens2M tok/h
Qwen: Qwen3 32Bqwen0,13 US$/M tokens2M tok/h
Meta: Llama 4 Maverickmeta-llama0,513 US$/M tokens4M tok/h
Meta: Llama 4 Scoutmeta-llama0,15 US$/M tokens37M tok/h
DeepSeek: DeepSeek V3 0324deepseek0,503 US$/M tokens38M tok/h
Mistral: Mistral Small 3.1 24Bmistralai0,402 US$/M tokens0.60M tok/h
Google: Gemma 3 4Bgoogle0,063 US$/M tokens0.43M tok/h
Google: Gemma 3 12Bgoogle0,075 US$/M tokens3M tok/h
Cohere: Command Acohere4,375 US$/M tokens1M tok/h
Reka Flash 3rekaai0,125 US$/M tokens2M tok/h
Google: Gemma 3 27Bgoogle0,148 US$/M tokens2M tok/h
TheDrummer: Skyfall 36B V2thedrummer0,613 US$/M tokens0.39M tok/h
AionLabs: Aion-RP 1.0 (8B)aion-labs1,00 US$/M tokens0.07M tok/h
Qwen: Qwen2.5 VL 72B Instructqwen0,85 US$/M tokens3M tok/h
Mistral: Mistral Small 3mistralai0,058 US$/M tokens4M tok/h
Microsoft: Phi 4microsoft0,088 US$/M tokens3M tok/h
DeepSeek: DeepSeek V3deepseek0,463 US$/M tokens41M tok/h
Cohere: Command R7B (12-2024)cohere0,066 US$/M tokens1M tok/h
Meta: Llama 3.3 70B Instructmeta-llama0,71 US$/M tokens3M tok/h
Mistral Large 2407mistralai3,00 US$/M tokens2M tok/h
Qwen2.5 Coder 32B Instructqwen0,745 US$/M tokens0.32M tok/h
TheDrummer: UnslopNemo 12Bthedrummer0,40 US$/M tokens0.60M tok/h
Magnum v4 72Banthracite-org3,125 US$/M tokens0.76M tok/h
Meta: Llama 3.2 1B Instructmeta-llama0,071 US$/M tokens0.38M tok/h
Meta: Llama 3.2 3B Instructmeta-llama0,12 US$/M tokens0.22M tok/h
Qwen2.5 72B Instructqwen0,37 US$/M tokens6M tok/h
Cohere: Command R (08-2024)cohere0,262 US$/M tokens0.92M tok/h
Cohere: Command R+ (08-2024)cohere4,375 US$/M tokens1M tok/h
Sao10K: Llama 3 8B Lunarissao10k0,042 US$/M tokens2M tok/h
Meta: Llama 3.1 70B Instructmeta-llama0,72 US$/M tokens3M tok/h
Meta: Llama 3.1 8B Instructmeta-llama0,025 US$/M tokens3M tok/h
Mistral: Mistral Nemomistralai0,029 US$/M tokens8M tok/h
Google: Gemma 2 27Bgoogle0,65 US$/M tokens0.37M tok/h
Mistral: Mixtral 8x22B Instructmistralai3,00 US$/M tokens0.24M tok/h

Ver todos los precios de GPU

¿API o GPU propia? — overfeed.news