Saltar al contenido
Alfa
  • Modelos
  • Clasificaciones
  • Ori
Hablar con ventas

Pie de página

OpenRouterOpenRouter
© 2026 OpenRouter, Inc

Producto

  • Chat
  • Clasificaciones
  • Pruebas de rendimiento
  • Aplicaciones
  • Descubrir
  • Modelos
  • Ori
  • Colecciones
  • Proveedores
  • Herramientas
  • Precios
  • Business
  • Enterprise
  • Laboratorio

Empresa

  • Acerca de
  • Blog
  • Clientes
  • Empleo
    Empleo
  • Privacidad
  • Términos del servicio
  • Centro de confianza
  • Soporte
  • Compatible con OR
  • Datos
  • Marca

Desarrollador

  • Documentación
  • Referencia de la API
  • Plataforma para desarrolladores
  • Estado
  • Mapa del sitio para IA

Conectar

  • Discord
  • GitHub
  • LinkedIn
  • X
  • YouTube
Favicon de openai

OpenAI: gpt-oss-20b

openai/gpt-oss-20b

Pesos del modelo
Comparar

gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for lower-latency inference and deployability on consumer or single-GPU hardware. The model is trained in OpenAI’s Harmony response format and supports reasoning level configuration, fine-tuning, and agentic capabilities including function calling, tool use, and structured outputs.

Modalidades
Precio de entrada / salida
0,006 US$ / 0,026 US$por 1M
Contexto
131K
Publicado
5 ago 2025
Fecha de corte del conocimiento
jun 2024
Comparar
ProveedoresPreciosRendimientoTiempo de actividadPruebas de rendimientoAplicacionesActividadPreguntas frecuentesExplorar

Proveedores

Distintas empresas alojan el mismo modelo. OpenRouter dirige su solicitud a una de ellas según el modo de enrutamiento que elija: Equilibrado (precio + velocidad), Nitro (el más rápido) o Floor (el más barato).

Precios

El precio medio que los clientes pagan realmente por este modelo, junto a los precios que publican los proveedores. El uso de caché y los descuentos hacen que el precio realmente pagado a menudo sea muy inferior al publicado.

Rendimiento

La velocidad de generación indica lo rápido que escribe el modelo (tokens por segundo: cuanto mayor, mejor). La latencia es el tiempo total de ida y vuelta (cuanto menor, mejor). TTFT es el tiempo hasta el primer token: cuánto tarda en aparecer algo en pantalla (cuanto menor, mejor).

Tiempo de actividad

El tiempo de actividad es el porcentaje de los últimos 3 días en que al menos un proveedor respondía a las solicitudes. La disponibilidad es el porcentaje de tiempo en que la inferencia se sirvió correctamente. OpenRouter supervisa continuamente y utiliza el siguiente mejor proveedor cuando uno devuelve un error.

Pruebas de rendimiento

Puntuaciones en evaluaciones estandarizadas y respuestas de este modelo a instrucciones compartidas de las pruebas de referencia. Los porcentajes más altos son mejores, y cada posición indica dónde se sitúa este modelo entre los modelos evaluados en esa fuente.

Resumen de puntuaciones en pruebas de rendimiento de OpenAI: gpt-oss-20b (Artificial Analysis, Design Arena y Vals AI)
FuentePrueba de rendimientoPuntuación
Artificial Analysisgpt-oss-20b (High) Coding Index20,7
Artificial Analysisgpt-oss-20b (High) Agentic Index1,2
Artificial Analysisgpt-oss-20b (High) GPQA Diamond68.8%
Artificial Analysisgpt-oss-20b (High) HLE11.0%
Artificial Analysisgpt-oss-20b (High) IFBench65.1%
Artificial Analysisgpt-oss-20b (High) τ²-Bench Telecom60.2%
Artificial Analysisgpt-oss-20b (High) AA-LCR34.7%
Artificial Analysisgpt-oss-20b (High) τ-Bench Banking7.0%
Artificial Analysisgpt-oss-20b (High) GDPval-AA0.0%
Artificial Analysisgpt-oss-20b (High) CritPt1.4%
Artificial Analysisgpt-oss-20b (High) LiveCodeBench77.7%
Artificial Analysisgpt-oss-20b (High) SciCode38.9%
Artificial Analysisgpt-oss-20b (High) Terminal-Bench Hard10.6%
Artificial Analysisgpt-oss-20b (High) Terminal-Bench 4.00.0%
Artificial Analysisgpt-oss-20b (High) Terminal-Bench 2.113.9%
Artificial Analysisgpt-oss-20b (High) MMLU-Pro74.8%
Artificial Analysisgpt-oss-20b (High) AIME 202589.3%
Artificial Analysisgpt-oss-20b (Low) GPQA Diamond61.1%
Artificial Analysisgpt-oss-20b (Low) HLE5.3%
Artificial Analysisgpt-oss-20b (Low) IFBench57.8%
Artificial Analysisgpt-oss-20b (Low) τ²-Bench Telecom50.3%
Artificial Analysisgpt-oss-20b (Low) AA-LCR31.0%
Artificial Analysisgpt-oss-20b (Low) CritPt0.0%
Artificial Analysisgpt-oss-20b (Low) LiveCodeBench65.2%
Artificial Analysisgpt-oss-20b (Low) Terminal-Bench Hard4.5%
Artificial Analysisgpt-oss-20b (Low) MMLU-Pro71.8%
Artificial Analysisgpt-oss-20b (Low) AIME 202562.3%
Design ArenaGPT OSS 20B Models Arena Data Visualization Elo933
Design ArenaGPT OSS 20B Models Arena Website Elo852
Vals AIGPT OSS 20B AIME86.0%
Vals AIGPT OSS 20B CaseLaw v243.8%
Vals AIGPT OSS 20B CorpFin v253.1%
Vals AIGPT OSS 20B GPQA Diamond68.9%
Vals AIGPT OSS 20B LiveCodeBench80.4%
Vals AIGPT OSS 20B LegalBench70.8%
Vals AIGPT OSS 20B MATH 50094.2%
Vals AIGPT OSS 20B MedQA82.9%
Vals AIGPT OSS 20B MGSM89.0%
Vals AIGPT OSS 20B MMLU Pro71.6%
Vals AIGPT OSS 20B TaxEval v263.7%

Aplicaciones

Aplicaciones públicas que envían más tráfico a este modelo. Una buena señal de cómo son las cargas de trabajo reales en producción, y una pista sobre los casos de uso para los que este modelo es más adecuado.

Actividad

Volumen de tokens y tráfico de solicitudes a este modelo a lo largo del tiempo.

API

Código listo para usar para llamar a este modelo. La API de OpenRouter es compatible con OpenAI: la mayoría de los SDK funcionan con solo cambiar la URL base. Lo único que cambia entre modelos es el identificador del modelo que aparece a continuación.

Disponible mediante enrutamiento dentro de la región de la UE

OpenAI: gpt-oss-20b está disponible mediante el enrutamiento dentro de la región de la UE. Las solicitudes enviadas a eu.openrouter.ai se descifran en la UE y se enrutan únicamente a endpoints compartidos de proveedores en Europa (la UE/el EEE, el Reino Unido y Suiza).

Proveedores en Europa
Amazon Bedrock
Precios en Europa
  • Amazon Bedrock: 0,07 US$/M tokens de entrada, 0,15 US$/M tokens de salida
URL base
https://eu.openrouter.ai/api/v1
Plan
Requiere Business o Enterprise. Autoservicio en Business: sin contrato ni mínimo mensual. OpenRouter Business para el enrutamiento dentro de la UE

Falla de forma cerrada: si no hay ningún endpoint de proveedor compartido disponible en Europa, la solicitud devuelve un error en lugar de salir de la región. Las solicitudes atendidas con su propia clave BYOK se dirigen a la región de su despliegue, que usted controla. Documentación de enrutamiento dentro de la región

Explorar más modelos

Clasificaciones de modelos de IAClasificación
Latencia / rendimiento
0,006 US$0,026 US$0,0050 US$0,45s115 tps
99,42 %
0,018 US$0,090 US$0,0090 US$1,55s56 tps
99,98 %
0,020 US$0,100 US$--0,42s130 tps
99,98 %
0,029 US$0,140 US$0,0290 US$0,56s57 tps
99,98 %
0,030 US$0,130 US$0,0300 US$0,30s124 tps
99,91 %
0,030 US$0,140 US$--0,85s64 tps
99,74 %
0,030 US$0,150 US$0,0200 US$1,82s97 tps
99,70 %
0,040 US$0,180 US$--1,06s41 tps
91,35 %
0,070 US$0,150 US$--0,41s419 tps
100,00 %
0,075 US$0,300 US$0,0375 US$0,38s380 tps
100,00 %
No se utiliza en el enrutamiento Standard:Por qué no se usan estos endpoints
0,070 US$0,150 US$--0,36s126 tps
100,00 %
Rendimiento
419tok/s

P50, el mejor entre los proveedores

Latencia
0,30s

P50, mejor proveedor

Pruebas de rendimiento de AutoExacto
GPQA DiamondTAU-BenchDekaLLM68,5 %--SiliconFlow63,0 %--auto-routing66,7 %52,5 %Novita58,1 %--Darkbloom64,7 %51,1 %
+8 proveedores más
Tiempo de actividad (3 d)El modelo estaba accesible. Solicitud enrutada a un proveedor.
100,00 %
Disponibilidad (3 d)El modelo devolvió una inferencia de cualquier proveedor. Los errores y las respuestas vacías cuentan en su contra.
97,36 %

Disponibilidad durante los últimos 3 días

Últimas 72 horas
Disponibilidad 97,36 %
Hace 3 díasHace 2 díasAyerAhora

Disponibilidad en las últimas 24 horas

OpenRouter Availability
97,10 %
Without Routing
93,96 %

Cuando se produce un error en un proveedor ascendente, podemos recuperarnos enrutando la solicitud a otro proveedor que funcione correctamente, si los filtros de su solicitud lo permiten. Puede acceder mediante programación a los datos de tiempo de actividad de cada proveedor a través de la API de endpoints. Obtenga más información sobre nuestras opciones de equilibrio de carga y enrutamiento de proveedores.

1.
Favicon de https://heyditto.ai/
HeyDitto
nuevo
137Btókenes
2.
Favicon de https://jobleads.com/
JobLeads LLM
Use AI to find jobs for you
31,2Btókenes
3.
Favicon de https://centralcommand.io/
Central Command
nuevo
22,8Btókenes
4.
Favicon de https://fvchat.app/
FVChat
nuevo
9,8Btókenes
5.
Favicon de https://gdevelop.io/
GDevelop
AI assisted game engine.
7,11Btókenes

Preguntas frecuentes

gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for lower-latency inference and deployability on consumer or single-GPU hardware.

gpt-oss-20b cuesta 0,006 US$/M de tokens de entrada y 0,026 US$/M de tókenes de salida, con tarifas independientes para Cache Read a 0,005 US$/M tokens.

gpt-oss-20b tiene una ventana de contexto de 131.072 tókenes. Admite hasta 32.768 tókenes de respuesta.

Sí. gpt-oss-20b acepta tools y tool_choice para llamadas a funciones en 9 de los 10 proveedores que lo ofrecen, y las solicitudes que envían herramientas se enrutan a esos proveedores. También admite salidas estructuradas mediante un esquema JSON en response_format.

gpt-oss-20b está disponible a través de 10 proveedores en OpenRouter: Prime Intellect, Darkbloom, AkashML, DekaLLM, CoreWeave, DeepInfra, Parasail, SiliconFlow y 2 más. Las solicitudes se enrutan al mejor proveedor disponible, con conmutación automática a los demás en caso de fallo, y puede fijar o excluir proveedores mediante el enrutamiento de proveedores.

gpt-oss-20b se lanzó el 5 de agosto de 2025. Su fecha límite de conocimientos es 30 de junio de 2024.

Más modelos de OpenAI

GPT-6.1 Terra Pro

GPT-6.1 Terra Pro is the same underlying model as GPT-6.1 Terra, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Terra and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Terra instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M1 US$ / 5 US$
GPT-6.1 Terra Pro (batch)

GPT-6.1 Terra Pro is the same underlying model as GPT-6.1 Terra, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Terra and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Terra instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M0,50 US$ / 2,50 US$
GPT-6.1 Terra

GPT-6.1 Terra is the mid-tier model in OpenAI's GPT-6.1 series, priced below GPT-6.1 Sol. It is suited for everyday agentic, coding, and writing work, is particularly strong at understanding screenshots and on-screen content, and is efficient enough for high-volume workloads.

TextoContexto de 1,1M1 US$ / 5 US$
GPT-6.1 Terra (batch)

GPT-6.1 Terra is the mid-tier model in OpenAI's GPT-6.1 series, priced below GPT-6.1 Sol. It is suited for everyday agentic, coding, and writing work, is particularly strong at understanding screenshots and on-screen content, and is efficient enough for high-volume workloads.

TextoContexto de 1,1M0,50 US$ / 2,50 US$
GPT-6 Luna Decisions

GPT-6 Luna Decisions is GPT-6 Luna served through OpenAI's Decisions API. Instead of generating text, it reads the content passed as state (text, JSON, or images) and returns typed, probabilistic answers to named questions in a single request: the probability of yes for a yes/no question (noul), a probability for every option plus the most likely one (choice), or a probability for every level of an ordered rubric plus the expected score (score).

It is built for classification, routing, moderation, and rubric grading where application code thresholds the returned numbers rather than parsing a chat reply. A request can carry up to 200 questions about the same content.

Decisiones0,10 US$ / 0 US$
GPT-6.1 Sol Pro

GPT-6.1 Sol Pro is the same underlying model as GPT-6.1 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M2 US$ / 10 US$
GPT-6.1 Sol Pro (batch)

GPT-6.1 Sol Pro is the same underlying model as GPT-6.1 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6.1 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6.1 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M1 US$ / 5 US$
GPT-6.1 Sol

GPT-6.1 Sol is an upgrade to GPT-6 Sol from OpenAI, positioned below the flagship GPT-6 Astra in the GPT-6 series. It is suited for agentic coding, computer use, document-heavy professional work, and multi-step business workflow automation, and approaches Astra-level results on these tasks at a much lower cost. Compared with GPT-6 Sol, it makes fewer factual errors and is more reliable at respecting explicit restrictions and user intent during agentic tasks.

TextoContexto de 1,1M2 US$ / 10 US$
GPT-6.1 Sol (batch)

GPT-6.1 Sol is an upgrade to GPT-6 Sol from OpenAI, positioned below the flagship GPT-6 Astra in the GPT-6 series. It is suited for agentic coding, computer use, document-heavy professional work, and multi-step business workflow automation, and approaches Astra-level results on these tasks at a much lower cost. Compared with GPT-6 Sol, it makes fewer factual errors and is more reliable at respecting explicit restrictions and user intent during agentic tasks.

TextoContexto de 1,1M1 US$ / 5 US$
GPT-6 Luna Pro

GPT-6 Luna Pro is the same underlying model as GPT-6 Luna, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Luna and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Luna instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M0,10 US$ / 0,50 US$
GPT-6 Luna Pro (batch)

GPT-6 Luna Pro is the same underlying model as GPT-6 Luna, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Luna and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Luna instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M0,05 US$ / 0,25 US$
GPT-6 Luna

GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol. It is suited for high-volume and latency-sensitive workloads such as chat, classification, and lightweight agentic tasks, and at higher reasoning effort it can take on complex software engineering and computer-use tasks that previously called for a Sol-tier model. It shares the GPT-6 family's gains in factual reliability and its clearer, more concise communication style.

TextoContexto de 1,1M0,10 US$ / 0,50 US$
GPT-6 Luna (batch)

GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol. It is suited for high-volume and latency-sensitive workloads such as chat, classification, and lightweight agentic tasks, and at higher reasoning effort it can take on complex software engineering and computer-use tasks that previously called for a Sol-tier model. It shares the GPT-6 family's gains in factual reliability and its clearer, more concise communication style.

TextoContexto de 1,1M0,05 US$ / 0,25 US$
GPT-6 Sol Pro

GPT-6 Sol Pro is the same underlying model as GPT-6 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M2 US$ / 10 US$
GPT-6 Sol Pro (batch)

GPT-6 Sol Pro is the same underlying model as GPT-6 Sol, served with reasoning.mode set to pro for higher-quality responses on complex tasks.

Cost note: pro mode spends far more reasoning tokens per request, so a typical request costs several times more than the same request on GPT-6 Sol and takes much longer to complete. It is intended for hard, high-stakes problems where the extra accuracy justifies the cost. For everyday coding, agentic, and chat workloads, use GPT-6 Sol instead.

Learn more in OpenAI's docs: https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode

TextoContexto de 1,1M1 US$ / 5 US$
GPT-6 Sol

GPT-6 Sol is the cost-efficient high-end model in OpenAI's GPT-6 series, positioned below the flagship GPT-6 Astra and above the fast GPT-6 Luna tier. It is suited for demanding professional work, agentic coding, business workflow automation, and computer use, and is particularly strong at long-horizon software engineering tasks in real codebases. It approaches Astra-level factual reliability at a much lower cost and shares Astra's clearer, more concise communication style in technical and coding conversations.

TextoContexto de 1,1M2 US$ / 10 US$
GPT-6 Sol (batch)

GPT-6 Sol is the cost-efficient high-end model in OpenAI's GPT-6 series, positioned below the flagship GPT-6 Astra and above the fast GPT-6 Luna tier. It is suited for demanding professional work, agentic coding, business workflow automation, and computer use, and is particularly strong at long-horizon software engineering tasks in real codebases. It approaches Astra-level factual reliability at a much lower cost and shares Astra's clearer, more concise communication style in technical and coding conversations.

TextoContexto de 1,1M1 US$ / 5 US$
GPT Astra Latest

This model always redirects to the latest model in the GPT Astra family.

TextoContexto de 1,1M
GPT Sol Latest

This model always redirects to the latest model in the GPT Sol family.

TextoContexto de 1,1M
GPT Terra Latest

This model always redirects to the latest model in the GPT Terra family.

TextoContexto de 1,1M

gpt-oss-20b cuesta 0,006 US/Mdetokensdeentraday0,026 US/M de tokens de entrada y 0,026 US/Mdetokensdeentraday0,026 US/M de tókenes de salida, con tarifas independientes para Cache Read a 0,005 US$/M tokens.