Artículo
Amazon Bedrock: Inteligencia vs Costo
El catálogo de Amazon Bedrock no para de expandirse: hoy cuenta con más de 115 modelos en us-east-1. Sin embargo, para cualquier equipo de ingeniería o producto la pregunta clave sigue siendo la misma: ¿cuánta inteligencia compras realmente por cada dólar invertido y en qué punto vale la pena pagar más?
Para responder a esto, este análisis combina los precios oficiales on-demand de Bedrock[1] con las métricas públicas del Intelligence Index de Artificial Analysis (v4.3.2)[2]. El estudio abarca 53 modelos activos evaluados en 107 configuraciones (desglosando cada nivel de esfuerzo de razonamiento disponible), con fecha de corte al 20 de septiembre de 2026.
Cómo leer el gráfico
El gráfico enfrenta el costo de entrada frente a la capacidad de cada modelo:
- Eje X (Costo): Precio por millón de tokens de entrada en escala logarítmica (USD, lista on-demand de Bedrock).
- Eje Y (Inteligencia): Puntuación en el Intelligence Index.
- Puntos y variantes: Cada punto representa un modelo. Cuando un modelo ofrece varios niveles de razonamiento (low, medium, high, max), se grafica un punto por cada nivel en la misma coordenada X: en Bedrock el costo por token de entrada no varía según el nivel de razonamiento, por lo que ese incremento vertical representa inteligencia adicional sin sobrecosto en el precio base de entrada.
- Frontera de Pareto: Al activar el interruptor, se traza la línea de eficiencia óptima. Conecta los modelos para los cuales no existe alternativa en el catálogo que sea simultáneamente más barata y más inteligente.
Usa la rueda o pellizca para hacer zoom · arrastra para explorar · doble clic para restablecer
Precio: tier estándar (≤272K tokens de input) de las fichas de modelo de Amazon Bedrock (docs.aws.amazon.com/bedrock/latest/userguide/model-cards.html), con la página general de pricing como respaldo cuando la ficha no trae su propia tabla, us-east-1. Índice de inteligencia: Artificial Analysis Intelligence Index v4.3.2 (artificialanalysis.ai). Cuando un modelo publica más de un nivel de esfuerzo de razonamiento, cada nivel es su propio punto — el precio de Bedrock no cambia con el esfuerzo, así que comparten x y se reparten en y. Un punto sobre la frontera no es dominado en precio ni en índice por ningún otro punto del set. Datos con corte al 20 de septiembre de 2026 — los catálogos de precio e índice cambian seguido, así que esto es una foto de ese día, no en vivo.
Top 10 por Intelligence Index
- 1Claude Fable 5.1 · max$1053.4
- 2Claude Fable 5.1 · xhigh$1053.2
- 3GPT-6 Astra · max$1152.7
- 4GPT-6 Astra · xhigh$1152.4
- 5Claude Fable 5.1 · high$1051.1
- 6GPT-6 Astra · high$1150.9
- 7Claude Opus 5 · max$550.8
- 8Claude Opus 5 · xhigh$549.7
- 9Claude Fable 5 · max$1049.6
- 10GPT-6 Astra · medium$1149.6
Ver los 107 puntos en tabla
| Modelo | Esfuerzo | Proveedor | $/1M input | Índice |
|---|---|---|---|---|
| Nova Micro | — | Amazon | $0.035 | 5.88 |
| NVIDIA Nemotron Nano 9B v2 | reasoning | NVIDIA | $0.06 | 7.43 |
| NVIDIA Nemotron Nano 9B v2 | non-reasoning | NVIDIA | $0.06 | 6.84 |
| Nova Lite | — | Amazon | $0.06 | 6.67 |
| GLM 4.7 Flash | reasoning | Z.AI | $0.07 | 14.87 |
| GLM 4.7 Flash | non-reasoning | Z.AI | $0.07 | 10.56 |
| gpt-oss-20b | low | OpenAI | $0.07 | 9.95 |
| gpt-oss-20b | high | OpenAI | $0.07 | 8.97 |
| gpt-oss-120b | high | OpenAI | $0.15 | 11.60 |
| Qwen3 Next 80B A3B | reasoning | Qwen | $0.15 | 11.20 |
| gpt-oss-120b | low | OpenAI | $0.15 | 10.21 |
| Qwen3-Coder-30B-A3B-Instruct | — | Qwen | $0.15 | 9.59 |
| Qwen3 32B (dense) | reasoning | Qwen | $0.15 | 8.59 |
| Qwen3 32B (dense) | non-reasoning | Qwen | $0.15 | 7.34 |
| Ministral 3 8B | — | Mistral AI | $0.15 | 5.48 |
| Mistral 7B Instruct | — | Mistral AI | $0.15 | 5.03 |
| Llama 4 Scout 17B Instruct | — | Meta | $0.17 | 8.08 |
| GPT-5.6 Luna | max | OpenAI | $0.22 | 37.32 |
| GPT-5.6 Luna | xhigh | OpenAI | $0.22 | 34.56 |
| GPT-5.6 Luna | high | OpenAI | $0.22 | 32.12 |
| GPT-5.6 Luna | medium | OpenAI | $0.22 | 25.04 |
| GPT-5.6 Luna | low | OpenAI | $0.22 | 21.01 |
| GPT-5.6 Luna | non-reasoning | OpenAI | $0.22 | 15.53 |
| Llama 4 Maverick 17B Instruct | — | Meta | $0.24 | 9.99 |
| MiniMax M2.5 | — | MiniMax | $0.3 | 22.80 |
| MiniMax M2.1 | — | MiniMax | $0.3 | 20.95 |
| MiniMax M2 | — | MiniMax | $0.3 | 18.63 |
| Nova 2 Lite | high | Amazon | $0.3 | 13.37 |
| Nova 2 Lite | medium | Amazon | $0.3 | 12.48 |
| Nova 2 Lite | low | Amazon | $0.3 | 11.80 |
| Nova 2 Lite | non-reasoning | Amazon | $0.3 | 8.74 |
| Mixtral 8x7B Instruct | — | Mistral AI | $0.45 | 5.12 |
| Mistral Large 3 | — | Mistral AI | $0.5 | 9.27 |
| Qwen3 Coder Next | — | Qwen | $0.5 | 9.24 |
| Qwen3 VL 235B A22B | reasoning | Qwen | $0.53 | 13.44 |
| Kimi K2.5 | reasoning | Moonshot AI | $0.6 | 23.46 |
| GLM 4.7 | reasoning | Z.AI | $0.6 | 22.24 |
| Kimi K2 Thinking | — | Moonshot AI | $0.6 | 22.02 |
| Kimi K2.5 | non-reasoning | Moonshot AI | $0.6 | 19.43 |
| GLM 4.7 | non-reasoning | Z.AI | $0.6 | 17.36 |
| DeepSeek V3.2 | reasoning | DeepSeek | $0.62 | 21.49 |
| DeepSeek V3.2 | non-reasoning | DeepSeek | $0.62 | 16.04 |
| Llama 3.3 70B Instruct | — | Meta | $0.72 | 7.66 |
| Llama 3.1 8B Instruct | — | Meta | $0.72 | 6.93 |
| Llama 3.1 70B Instruct | — | Meta | $0.72 | 6.60 |
| Nova Pro | — | Amazon | $0.8 | 6.96 |
| GLM 5 | reasoning | Z.AI | $1 | 27.91 |
| GLM 5 | non-reasoning | Z.AI | $1 | 21.78 |
| Claude Haiku 4.5 | reasoning | Anthropic | $1 | 16.88 |
| Claude Haiku 4.5 | non-reasoning | Anthropic | $1 | 15.41 |
| Mistral Small (24.02) | — | Mistral AI | $1 | 5.85 |
| DeepSeek-R1 | — | DeepSeek | $1.35 | 11.41 |
| Claude Sonnet 5 | max | Anthropic | $2 | 38.16 |
| Claude Sonnet 5 | xhigh | Anthropic | $2 | 34.38 |
| Claude Sonnet 5 | high | Anthropic | $2 | 31.66 |
| Claude Sonnet 5 | medium | Anthropic | $2 | 28.05 |
| Claude Sonnet 5 | low | Anthropic | $2 | 24.26 |
| Claude Sonnet 5 | non-reasoning | Anthropic | $2 | 23.20 |
| Pixtral Large (25.02) | — | Mistral AI | $2 | 7.15 |
| Grok 4.6 | high | xAI | $2.2 | 44.31 |
| Grok 4.6 | xhigh | xAI | $2.2 | 44.20 |
| Grok 4.6 | medium | xAI | $2.2 | 42.84 |
| GPT-5.6 Terra | max | OpenAI | $2.2 | 42.08 |
| GPT-5.6 Terra | xhigh | OpenAI | $2.2 | 37.95 |
| Grok 4.6 | low | xAI | $2.2 | 35.12 |
| GPT-5.6 Terra | high | OpenAI | $2.2 | 34.24 |
| GPT-5.6 Terra | medium | OpenAI | $2.2 | 30.09 |
| GPT-5.6 Terra | low | OpenAI | $2.2 | 27.50 |
| GPT-5.6 Terra | non-reasoning | OpenAI | $2.2 | 20.78 |
| Llama 3 70B Instruct | — | Meta | $2.65 | 5.45 |
| Llama 3 8B Instruct | — | Meta | $2.65 | 4.83 |
| Claude Sonnet 4.6 | max | Anthropic | $3 | 30.06 |
| Claude Sonnet 4.6 | non-reasoning | Anthropic | $3 | 23.30 |
| Claude Sonnet 4.5 | reasoning | Anthropic | $3 | 20.67 |
| Claude Sonnet 4.5 | non-reasoning | Anthropic | $3 | 19.34 |
| Kimi K3 | max | Moonshot AI | $3.3 | 43.59 |
| Kimi K3 | low | Moonshot AI | $3.3 | 34.46 |
| Mistral Large (24.02) | — | Mistral AI | $4 | 5.76 |
| GPT-5.6 Sol | max | OpenAI | $4.4 | 46.97 |
| GPT-5.6 Sol | xhigh | OpenAI | $4.4 | 44.01 |
| GPT-5.6 Sol | high | OpenAI | $4.4 | 42.35 |
| GPT-5.6 Sol | medium | OpenAI | $4.4 | 39.24 |
| GPT-5.6 Sol | low | OpenAI | $4.4 | 33.47 |
| GPT-5.6 Sol | non-reasoning | OpenAI | $4.4 | 28.33 |
| Claude Opus 5 | max | Anthropic | $5 | 50.78 |
| Claude Opus 5 | xhigh | Anthropic | $5 | 49.68 |
| Claude Opus 5 | high | Anthropic | $5 | 48.12 |
| Claude Opus 5 | medium | Anthropic | $5 | 44.83 |
| Claude Opus 4.8 | max | Anthropic | $5 | 41.79 |
| Claude Opus 4.7 | max | Anthropic | $5 | 40.69 |
| Claude Opus 5 | low | Anthropic | $5 | 39.35 |
| Claude Opus 4.6 | max | Anthropic | $5 | 31.95 |
| Claude Opus 4.7 | non-reasoning | Anthropic | $5 | 30.93 |
| Claude Opus 4.5 | reasoning | Anthropic | $5 | 29.10 |
| Claude Opus 4.6 | non-reasoning | Anthropic | $5 | 26.35 |
| Claude Opus 4.5 | non-reasoning | Anthropic | $5 | 23.68 |
| Claude Fable 5.1 | max | Anthropic | $10 | 53.35 |
| Claude Fable 5.1 | xhigh | Anthropic | $10 | 53.20 |
| Claude Fable 5.1 | high | Anthropic | $10 | 51.15 |
| Claude Fable 5 | max | Anthropic | $10 | 49.63 |
| Claude Fable 5.1 | medium | Anthropic | $10 | 48.92 |
| Claude Fable 5.1 | low | Anthropic | $10 | 46.82 |
| GPT-6 Astra | max | OpenAI | $11 | 52.67 |
| GPT-6 Astra | xhigh | OpenAI | $11 | 52.39 |
| GPT-6 Astra | high | OpenAI | $11 | 50.92 |
| GPT-6 Astra | medium | OpenAI | $11 | 49.57 |
| GPT-6 Astra | low | OpenAI | $11 | 45.78 |
Hallazgos clave en la frontera
De las 107 configuraciones analizadas, solo 9 se sitúan sobre la frontera de Pareto. De este grupo se desprenden tres conclusiones contundentes:
- El esfuerzo de razonamiento aporta más que migrar de modelo. Configurar GPT-5.6 Luna en esfuerzo max alcanza un índice de 37.3 por $0.22/1M tokens —prácticamente a la par de Claude Sonnet 5 en esfuerzo max (índice 38.2), pero a una novena parte del costo ($2.00/1M). La misma versión de Luna sin razonamiento (non-reasoning) apenas marca 15.5: la ganancia al activar el razonamiento dentro de un mismo modelo supera la brecha que existe entre proveedores completos al mismo rango de precio.
- Por debajo de $1.00/1M, dominan las opciones abiertas. Amazon Nova Micro, NVIDIA Nemotron Nano 9B v2 y GLM 4.7 Flash dominan el segmento de bajo costo; nadie ofrece más inteligencia por menos dinero en esa franja. A partir de $0.22/1M, la frontera pasa a manos de OpenAI, Anthropic y xAI, casi siempre en sus niveles máximos de razonamiento (high o max).
- Rendimientos decrecientes en la gama alta. A partir de la barrera de los $2.20/1M (Grok 4.6, GPT-5.6 Terra, Claude Opus 5 y Claude Fable 5.1), multiplicar el precio hasta por 5x ofrece mejoras incrementales relativamente modestas en el índice de inteligencia, delimitando con claridad el punto en que el costo adicional solo se justifica para tareas altamente críticas.
De los benchmarks a producción: ¿por qué modelos empezar?
Ningún benchmark sintético reemplaza las pruebas con tus propios datos y flujos de trabajo. En producción, la métrica definitiva no es el precio por millón de tokens, sino el costo por tarea exitosa:
- Lo “barato” puede salir caro: Un modelo económico que falla con frecuencia, requiere reintentos constantes o prompts excesivamente largos para no alucinar, termina costando más tiempo y dinero.
- Lo más potente suele ser innecesario: Usar el modelo más costoso para tareas repetitivas, clasificación o extracción estructurada es pagar de más sin beneficio real.
Cómo usar esta guía: Probar decenas de modelos contra tus propios datos no es viable. Utiliza la frontera de Pareto como tu shortlist de partida: identifica los 2 o 3 modelos que dominan la relación costo-inteligencia para el nivel de complejidad que necesitas y empieza tus pruebas ahí. Si el más eficiente cumple con el estándar de calidad de tu caso de uso, ya encontraste tu ganador; si no, escala al siguiente escalón de la frontera.
Explora los 77 modelos activos
El gráfico superior se limita a modelos con precio e índice reportados. La siguiente tabla interactiva reúne los 77 modelos en estado ACTIVE en Bedrock[3] que cuentan con precio por token o benchmark (excluyendo aquellos con tarifas por imagen, segundo de video o consulta). Puedes filtrar por proveedor, soporte multimodal (imágenes), ventana de contexto y tokens máximos de salida para evaluar la alternativa ideal para tu arquitectura:
| Modelo | Proveedor | Esfuerzo | Contexto | $/1M in | Índice ↓ |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic | max | 1M | $10 | 53.4 |
| Claude Fable 5.1 | Anthropic | xhigh | 1M | $10 | 53.2 |
| GPT-6 Astra | OpenAI | max | 272K(≤272K) | $11 | 52.7 |
| GPT-6 Astra | OpenAI | max | 1.1M(>272K) | $22 | 52.7 |
| GPT-6 Astra | OpenAI | xhigh | 272K(≤272K) | $11 | 52.4 |
| GPT-6 Astra | OpenAI | xhigh | 1.1M(>272K) | $22 | 52.4 |
| Claude Fable 5.1 | Anthropic | high | 1M | $10 | 51.1 |
| GPT-6 Astra | OpenAI | high | 272K(≤272K) | $11 | 50.9 |
| GPT-6 Astra | OpenAI | high | 1.1M(>272K) | $22 | 50.9 |
| Claude Opus 5 | Anthropic | max | 1M | $5 | 50.8 |
| Claude Opus 5 | Anthropic | xhigh | 1M | $5 | 49.7 |
| Claude Fable 5 | Anthropic | max | 1M | $10 | 49.6 |
| GPT-6 Astra | OpenAI | medium | 272K(≤272K) | $11 | 49.6 |
| GPT-6 Astra | OpenAI | medium | 1.1M(>272K) | $22 | 49.6 |
| Claude Fable 5.1 | Anthropic | medium | 1M | $10 | 48.9 |
| Claude Opus 5 | Anthropic | high | 1M | $5 | 48.1 |
| GPT-5.6 Sol | OpenAI | max | 272K(≤272K) | $4.4 | 47.0 |
| GPT-5.6 Sol | OpenAI | max | 1M(>272K) | $8.8 | 47.0 |
| Claude Fable 5.1 | Anthropic | low | 1M | $10 | 46.8 |
| GPT-6 Astra | OpenAI | low | 272K(≤272K) | $11 | 45.8 |
| GPT-6 Astra | OpenAI | low | 1.1M(>272K) | $22 | 45.8 |
| Claude Opus 5 | Anthropic | medium | 1M | $5 | 44.8 |
| Grok 4.6 | xAI | high | 500K | $2.2 | 44.3 |
| Grok 4.6 | xAI | xhigh | 500K | $2.2 | 44.2 |
| GPT-5.6 Sol | OpenAI | xhigh | 272K(≤272K) | $4.4 | 44.0 |
| GPT-5.6 Sol | OpenAI | xhigh | 1M(>272K) | $8.8 | 44.0 |
| Kimi K3 | Moonshot AI | max | 1M | $3.3 | 43.6 |
| Grok 4.6 | xAI | medium | 500K | $2.2 | 42.8 |
| GPT-5.6 Sol | OpenAI | high | 272K(≤272K) | $4.4 | 42.4 |
| GPT-5.6 Sol | OpenAI | high | 1M(>272K) | $8.8 | 42.4 |
| GPT-5.6 Terra | OpenAI | max | 272K(≤272K) | $2.2 | 42.1 |
| GPT-5.6 Terra | OpenAI | max | 1M(>272K) | $4.4 | 42.1 |
| Claude Opus 4.8 | Anthropic | max | 1M | $5 | 41.8 |
| Claude Opus 4.7 | Anthropic | max | 1M | $5 | 40.7 |
| Claude Opus 5 | Anthropic | low | 1M | $5 | 39.4 |
| GPT-5.6 Sol | OpenAI | medium | 272K(≤272K) | $4.4 | 39.2 |
| GPT-5.6 Sol | OpenAI | medium | 1M(>272K) | $8.8 | 39.2 |
| Claude Sonnet 5 | Anthropic | max | 1M | $2 | 38.2 |
| GPT-5.6 Terra | OpenAI | xhigh | 272K(≤272K) | $2.2 | 38.0 |
| GPT-5.6 Terra | OpenAI | xhigh | 1M(>272K) | $4.4 | 38.0 |
| GPT-5.6 Luna | OpenAI | max | 272K(≤272K) | $0.22 | 37.3 |
| GPT-5.6 Luna | OpenAI | max | 1M(>272K) | $0.44 | 37.3 |
| Grok 4.6 | xAI | low | 500K | $2.2 | 35.1 |
| GPT-5.6 Luna | OpenAI | xhigh | 272K(≤272K) | $0.22 | 34.6 |
| GPT-5.6 Luna | OpenAI | xhigh | 1M(>272K) | $0.44 | 34.6 |
| Kimi K3 | Moonshot AI | low | 1M | $3.3 | 34.5 |
| Claude Sonnet 5 | Anthropic | xhigh | 1M | $2 | 34.4 |
| GPT-5.6 Terra | OpenAI | high | 272K(≤272K) | $2.2 | 34.2 |
| GPT-5.6 Terra | OpenAI | high | 1M(>272K) | $4.4 | 34.2 |
| GPT-5.6 Sol | OpenAI | low | 272K(≤272K) | $4.4 | 33.5 |
| GPT-5.6 Sol | OpenAI | low | 1M(>272K) | $8.8 | 33.5 |
| GPT-5.6 Luna | OpenAI | high | 272K(≤272K) | $0.22 | 32.1 |
| GPT-5.6 Luna | OpenAI | high | 1M(>272K) | $0.44 | 32.1 |
| Claude Opus 4.6 | Anthropic | max | 1M | $5 | 31.9 |
| Claude Sonnet 5 | Anthropic | high | 1M | $2 | 31.7 |
| Claude Opus 4.7 | Anthropic | non-reasoning | 1M | $5 | 30.9 |
| GPT-5.6 Terra | OpenAI | medium | 272K(≤272K) | $2.2 | 30.1 |
| GPT-5.6 Terra | OpenAI | medium | 1M(>272K) | $4.4 | 30.1 |
| Claude Sonnet 4.6 | Anthropic | max | 1M | $3 | 30.1 |
| Claude Opus 4.5 | Anthropic | reasoning | 200K | $5 | 29.1 |
| GPT-5.6 Sol | OpenAI | non-reasoning | 272K(≤272K) | $4.4 | 28.3 |
| GPT-5.6 Sol | OpenAI | non-reasoning | 1M(>272K) | $8.8 | 28.3 |
| Claude Sonnet 5 | Anthropic | medium | 1M | $2 | 28.1 |
| GLM 5 | Z.AI | reasoning | 200K | $1 | 27.9 |
| GPT-5.6 Terra | OpenAI | low | 272K(≤272K) | $2.2 | 27.5 |
| GPT-5.6 Terra | OpenAI | low | 1M(>272K) | $4.4 | 27.5 |
| Claude Opus 4.6 | Anthropic | non-reasoning | 1M | $5 | 26.4 |
| GPT-5.6 Luna | OpenAI | medium | 272K(≤272K) | $0.22 | 25.0 |
| GPT-5.6 Luna | OpenAI | medium | 1M(>272K) | $0.44 | 25.0 |
| Claude Sonnet 5 | Anthropic | low | 1M | $2 | 24.3 |
| Claude Opus 4.5 | Anthropic | non-reasoning | 200K | $5 | 23.7 |
| Kimi K2.5 | Moonshot AI | reasoning | 256K | $0.6 | 23.5 |
| Claude Sonnet 4.6 | Anthropic | non-reasoning | 1M | $3 | 23.3 |
| Claude Sonnet 5 | Anthropic | non-reasoning | 1M | $2 | 23.2 |
| MiniMax M2.5 | MiniMax | — | 196K | $0.3 | 22.8 |
| GLM 4.7 | Z.AI | reasoning | 203K | $0.6 | 22.2 |
| Kimi K2 Thinking | Moonshot AI | — | 256K | $0.6 | 22.0 |
| GLM 5 | Z.AI | non-reasoning | 200K | $1 | 21.8 |
| DeepSeek V3.2 | DeepSeek | reasoning | 164K | $0.62 | 21.5 |
| GPT-5.6 Luna | OpenAI | low | 272K(≤272K) | $0.22 | 21.0 |
| GPT-5.6 Luna | OpenAI | low | 1M(>272K) | $0.44 | 21.0 |
| MiniMax M2.1 | MiniMax | — | 196K | $0.3 | 20.9 |
| GPT-5.6 Terra | OpenAI | non-reasoning | 272K(≤272K) | $2.2 | 20.8 |
| GPT-5.6 Terra | OpenAI | non-reasoning | 1M(>272K) | $4.4 | 20.8 |
| Claude Sonnet 4.5 | Anthropic | reasoning | 200K | $3 | 20.7 |
| Kimi K2.5 | Moonshot AI | non-reasoning | 256K | $0.6 | 19.4 |
| Claude Sonnet 4.5 | Anthropic | non-reasoning | 200K | $3 | 19.3 |
| MiniMax M2 | MiniMax | — | 1M | $0.3 | 18.6 |
| GLM 4.7 | Z.AI | non-reasoning | 203K | $0.6 | 17.4 |
| Claude Haiku 4.5 | Anthropic | reasoning | 200K | $1 | 16.9 |
| DeepSeek V3.2 | DeepSeek | non-reasoning | 164K | $0.62 | 16.0 |
| GPT-5.6 Luna | OpenAI | non-reasoning | 272K(≤272K) | $0.22 | 15.5 |
| GPT-5.6 Luna | OpenAI | non-reasoning | 1M(>272K) | $0.44 | 15.5 |
| Claude Haiku 4.5 | Anthropic | non-reasoning | 200K | $1 | 15.4 |
| GLM 4.7 Flash | Z.AI | reasoning | 203K | $0.07 | 14.9 |
| Qwen3 VL 235B A22B | Qwen | reasoning | 256K | $0.53 | 13.4 |
| Nova 2 Lite | Amazon | high | 1M | $0.3 | 13.4 |
| NVIDIA Nemotron 3 Super 120B A12B | NVIDIA | — | 256K | $0.15 | 12.8 |
| Nova 2 Lite | Amazon | medium | 1M | $0.3 | 12.5 |
| Nova 2 Lite | Amazon | low | 1M | $0.3 | 11.8 |
| gpt-oss-120b | OpenAI | high | 128K | $0.15 | 11.6 |
| DeepSeek-R1 | DeepSeek | — | 128K | $1.35 | 11.4 |
| Qwen3 Next 80B A3B | Qwen | reasoning | 256K | $0.15 | 11.2 |
| GLM 4.7 Flash | Z.AI | non-reasoning | 203K | $0.07 | 10.6 |
| gpt-oss-120b | OpenAI | low | 128K | $0.15 | 10.2 |
| Llama 4 Maverick 17B Instruct | Meta | — | 1M | $0.24 | 10.0 |
| gpt-oss-20b | OpenAI | low | 128K | $0.07 | 9.9 |
| Qwen3-Coder-30B-A3B-Instruct | Qwen | — | 256K | $0.15 | 9.6 |
| Mistral Large 3 | Mistral AI | — | 256K | $0.5 | 9.3 |
| Qwen3 Coder Next | Qwen | — | 256K | $0.5 | 9.2 |
| gpt-oss-20b | OpenAI | high | 128K | $0.07 | 9.0 |
| Nova 2 Lite | Amazon | non-reasoning | 1M | $0.3 | 8.7 |
| Devstral 2 123B | Mistral AI | — | 256K | $0.4 | 8.6 |
| Magistral Small 2509 | Mistral AI | — | 128K | $0.5 | 8.6 |
| Qwen3 32B (dense) | Qwen | reasoning | 32K | $0.15 | 8.6 |
| Llama 4 Scout 17B Instruct | Meta | — | 10M | $0.17 | 8.1 |
| Llama 3.3 70B Instruct | Meta | — | 128K | $0.72 | 7.7 |
| NVIDIA Nemotron Nano 9B v2 | NVIDIA | reasoning | 128K | $0.06 | 7.4 |
| Qwen3 32B (dense) | Qwen | non-reasoning | 32K | $0.15 | 7.3 |
| Pixtral Large (25.02) | Mistral AI | — | 128K | $2 | 7.2 |
| Nova Pro | Amazon | — | 300K | $0.8 | 7.0 |
| Llama 3.1 8B Instruct | Meta | — | 128K | $0.72 | 6.9 |
| Nemotron Nano 3 30B | NVIDIA | — | 256K | $0.06 | 6.8 |
| NVIDIA Nemotron Nano 9B v2 | NVIDIA | non-reasoning | 128K | $0.06 | 6.8 |
| Nova Lite | Amazon | — | 300K | $0.06 | 6.7 |
| Llama 3.1 70B Instruct | Meta | — | 128K | $0.72 | 6.6 |
| Ministral 14B 3.0 | Mistral AI | — | 128K | $0.2 | 6.0 |
| Nova Micro | Amazon | — | 128K | $0.035 | 5.9 |
| Mistral Small (24.02) | Mistral AI | — | 32K | $1 | 5.8 |
| NVIDIA Nemotron Nano 12B v2 VL BF16 | NVIDIA | — | 128K | $0.2 | 5.8 |
| Mistral Large (24.02) | Mistral AI | — | 32K | $4 | 5.8 |
| Ministral 3 8B | Mistral AI | — | 128K | $0.15 | 5.5 |
| Llama 3 70B Instruct | Meta | — | 8K | $2.65 | 5.5 |
| Mixtral 8x7B Instruct | Mistral AI | — | 32K | $0.45 | 5.1 |
| Mistral 7B Instruct | Mistral AI | — | 32K | $0.15 | 5.0 |
| Gemma 3 27B PT | — | 128K | $0.23 | 4.8 | |
| Ministral 3B | Mistral AI | — | 128K | $0.1 | 4.8 |
| Gemma 3 4B IT | — | 128K | $0.04 | 4.8 | |
| Llama 3 8B Instruct | Meta | — | 8K | $2.65 | 4.8 |
| Gemma 3 12B IT | — | 128K | $0.09 | 3.8 | |
| Nova 2 Sonic | Amazon | — | 1M | $3 | — |
| Titan Text Embeddings v2 | Amazon | — | 8K | $0.02 | — |
| Titan Embeddings G1 - Text | Amazon | — | 8K | $0.1 | — |
| Titan Text Embeddings V2 | Amazon | — | 8K | $0.02 | — |
| Embed English | Cohere | — | 512 | $0.1 | — |
| Embed Multilingual | Cohere | — | 512 | $0.1 | — |
| Embed v4 | Cohere | — | 128K | $0.12 | — |
| Voxtral Mini 3B 2507 | Mistral AI | — | 32K | $0.04 | — |
| Voxtral Small 24B 2507 | Mistral AI | — | 32K | $0.1 | — |
| GPT OSS Safeguard 120B | OpenAI | — | 128K | $0.15 | — |
| GPT OSS Safeguard 20B | OpenAI | — | 128K | $0.07 | — |
| Writer Palmyra Vision 7B | Writer | — | 4K | $0.15 | — |
| Palmyra X4 | Writer | — | 128K | $2.5 | — |
| Palmyra X5 | Writer | — | 128K | $0.6 | — |
154 de 154 filas · un modelo aparece varias veces si tiene distintos niveles de esfuerzo de razonamiento o distintos tiers de precio por context window · clic en un encabezado para ordenar


