Cloudflare: inteligencia, rendimiento y precio de sus modelos

Cloudflare
Cloudflare

Análisis de los modelos de Cloudflare en métricas clave como calidad, precio, velocidad de salida, latencia, ventana de contexto y más. Este análisis está pensado para ayudarte a elegir el mejor modelo ofrecido por Cloudflare para tu caso de uso.

Más inteligente

#1
Kimi K2.6
Kimi K2.6
44
#2
Gemma 4 26B A4B
Gemma 4 26B A4B
26
#3
gpt-oss-120b (high)
gpt-oss-120b (high)
24
#4
gpt-oss-120b (low)
gpt-oss-120b (low)
15
#5
gpt-oss-20b (high)
gpt-oss-20b (high)
15

Índice de inteligencia

11 modelos en total

Más rápido

#1
Llama 3.1 8B
Llama 3.1 8B
188 t/s
#2
gpt-oss-20b (high)
gpt-oss-20b (high)
165 t/s
#3
gpt-oss-20b (low)
gpt-oss-20b (low)
162 t/s
#4
gpt-oss-120b (high)
gpt-oss-120b (high)
116 t/s
#5
gpt-oss-120b (low)
gpt-oss-120b (low)
112 t/s

Velocidad de salida

11 modelos en total

Menor precio

#1
Gemma 4 26B A4B
Gemma 4 26B A4B
$0.12
#2
gpt-oss-20b (high)
gpt-oss-20b (high)
$0.21
#3
gpt-oss-20b (low)
gpt-oss-20b (low)
$0.21
#4
Llama 4 Scout
Llama 4 Scout
$0.33
#5
Mistral Small 3.1
Mistral Small 3.1
$0.37

Precio combinado (por 1M de tokens)

11 modelos en total

Indica un modelo de razonamiento

Cloudflare ofrece 11 modelos, cada uno con distintas características de inteligencia, rendimiento y precio. A continuación se comparan las métricas clave entre modelos.

  • En inteligencia, los mejores modelos en Cloudflare son Kimi K2.6 (44), Gemma 4 26B A4B (26) y gpt-oss-120b (high) (24).
  • En velocidad de salida, los modelos más rápidos son Llama 3.1 8B (188 t/s), gpt-oss-20b (high) (165 t/s) y gpt-oss-20b (low) (162 t/s). La velocidad varía significativamente entre modelos, con una diferencia del 68% entre el más rápido y el más lento.
  • En latencia, Llama 3.1 8B (0.76s), Llama 4 Scout (0.78s) y Mistral Small 3.1 (1.70s) ofrecen el menor tiempo hasta el primer token de respuesta.
  • En precios, Gemma 4 26B A4B ($0.12), gpt-oss-20b (high) ($0.21) y gpt-oss-20b (low) ($0.21) ofrecen los menores precios combinados por 1M de tokens. Los precios varían hasta 3.1x entre modelos.
  • En tamaño de ventana de contexto, Kimi K2.6 (262k), Gemma 4 26B A4B (256k) y Llama 4 Scout (131k) admiten las ventanas de contexto más grandes en Cloudflare.

Aspectos destacados

Inteligencia

Artificial Analysis Intelligence Index · Higher is better

Velocidad

Output tokens per second · Higher is better

Precio

USD per 1M tokens (blended) · Lower is better

Evaluaciones de inteligencia

Artificial Analysis Intelligence Index

Artificial Analysis Intelligence Index v4.1 incorporates 9 evaluations: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR
Estimate (independent evaluation forthcoming)
Reasoning models are indicated by a lightbulb icon

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Intelligence Evaluations

Intelligence evaluations measured independently by Artificial Analysis · Higher is better
GDPval-AA v2

Agentic real-world work tasks, (Elo-500)/2000

𝜏³-Banking

Agentic tool use

Terminal-Bench v2.1

Agentic coding & terminal use

SciCode

Coding

Humanity's Last Exam

Reasoning & knowledge

GPQA Diamond

Scientific reasoning

CritPt

Physics reasoning

AA-Omniscience Accuracy

Knowledge

AA-Omniscience Non-Hallucination Rate

1 - hallucination rate

AA-LCR

Long context reasoning

AA-Briefcase

Agentic knowledge work, Elo

AutomationBench-AA

Agentic SaaS workflows

Harvey LAB-AA

Legal agentic work, criterion pass rate

EnterpriseOps-Gym-AANew

Agentic business operations

IFBench

Instruction following

APEX-Agents-AA

Long-horizon agentic tasks

ITBench-AA

Kubernetes incident root-cause analysis

MMMU-Pro

Visual reasoning

Reasoning models are indicated by a lightbulb icon

While model intelligence generally translates across use cases, specific evaluations may be more relevant for certain use cases.

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Intelligence Index vs. Price

Blended at 7:2:1 (cache-input-output) · USD per 1M tokens (blended)
Most attractive quadrant
Reasoning models are indicated by a lightbulb icon

While higher intelligence models are typically more expensive, they do not all follow the same price-quality curve.

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

The blended cache price shown here uses cache hit price only. Other caching costs differ by provider:

  • Anthropic: charges a separate cache write fee, with different rates for 5-minute and 1-hour TTLs (1-hour TTL is more expensive).
  • Google (Vertex/Gemini): charges a per-hour cache storage fee in addition to cache hit pricing. Some providers also use tiered pricing for prompts above 200K tokens.
  • OpenAI, DeepSeek, others: typically charge only cache hit pricing with no write or storage fee.

See Prompt Caching for the full breakdown.

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Ventana de contexto

Context Window

Context window: tokens limit · Higher is better
Reasoning models are indicated by a lightbulb icon

Larger context windows are relevant to RAG (Retrieval Augmented Generation) LLM workflows which typically involve reasoning and information retrieval of large amounts of data.

Maximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).

Precios

Intelligence Index vs. Price

Blended at 7:2:1 (cache-input-output) · USD per 1M tokens (blended)
Most attractive quadrant
Reasoning models are indicated by a lightbulb icon

While higher intelligence models are typically more expensive, they do not all follow the same price-quality curve.

Artificial Analysis Intelligence Index v4.1 includes: GDPval-AA v2, 𝜏³-Banking, Terminal-Bench v2.1, SciCode, Humanity's Last Exam, GPQA Diamond, CritPt, AA-Omniscience, AA-LCR. See Intelligence Index methodology for further details, including a breakdown of each evaluation and how we run them.

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

The blended cache price shown here uses cache hit price only. Other caching costs differ by provider:

  • Anthropic: charges a separate cache write fee, with different rates for 5-minute and 1-hour TTLs (1-hour TTL is more expensive).
  • Google (Vertex/Gemini): charges a per-hour cache storage fee in addition to cache hit pricing. Some providers also use tiered pricing for prompts above 200K tokens.
  • OpenAI, DeepSeek, others: typically charge only cache hit pricing with no write or storage fee.

See Prompt Caching for the full breakdown.

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Resumen de rendimiento

Output Speed vs. Price

Output speed: output tokens per second · USD per 1M tokens (blended)
Most attractive quadrant
Reasoning models are indicated by a lightbulb icon

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

Velocidad

Medida por la velocidad de salida (tokens por segundo)

Output Speed

Output tokens per second · Higher is better
Reasoning models are indicated by a lightbulb icon

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Figures represent performance of the model's first-party API (e.g. OpenAI for o1) or the median across providers where a first-party API is not available (e.g. Meta's Llama models).

Latencia

Medida por el tiempo (segundos) hasta el primer token

Latency: Time To First Answer Token

Seconds to first answer token received · Accounts for reasoning model 'thinking' time
Reasoning models are indicated by a lightbulb icon

Time to first answer token received, in seconds, after API request sent. For reasoning models, this includes the 'thinking' time of the model before providing an answer. For models which do not support streaming, this represents time to receive the completion.

Tiempo de respuesta de extremo a extremo

Seconds to output 500 tokens, calculated based on time to first token, 'thinking' time for reasoning models, and output speed

End-to-End Response Time vs. Price

End-to-end response time: end-to-end seconds to output 500 tokens · USD per 1M tokens (blended)
Most attractive quadrant
Reasoning models are indicated by a lightbulb icon

Price per token, shown in USD per million tokens. Price is a blend of cache hit, input, and output token prices using the selected ratio (default 7:2:1 cache-input-output).

Seconds to receive a 500 token response. Key components:

  • Input time: Time to receive the first response token
  • Thinking time (only for reasoning models): Time reasoning models spend outputting tokens to reason prior to providing an answer. Amount of tokens based on the average reasoning tokens across a diverse set of 60 prompts (methodology details).
  • Answer time: Time to generate 500 output tokens, based on output speed

Figures represent median (P50) measurement over the past 72 hours to reflect sustained changes in performance.

Análisis adicional
Logo de Kimi
Kimi K2.6
262k
Abierto
44
$0.28
45
1.48
112.36
99.69
Modelo
Proveedor
Logo de Google
Gemma 4 26B A4B
256k
Abierto
26
$0.03
77
1.18
33.76
26.07
Modelo
Proveedor
Logo de OpenAI
gpt-oss-120b (high)
116k
Abierto
24
$0.12
116
0.93
22.42
17.19
Modelo
Proveedor
Logo de OpenAI
gpt-oss-120b (low)
128k
Abierto
15
$0.04
112
1.14
23.52
17.90
Modelo
Proveedor
Logo de OpenAI
gpt-oss-20b (high)
128k
Abierto
15
$0.06
165
0.78
15.89
12.09
Modelo
Proveedor
Logo de Mistral
Mistral Small 3.1
128k
Abierto
15
$0.14
44
1.70
13.06
--
Modelo
Proveedor
Logo de OpenAI
gpt-oss-20b (low)
128k
Abierto
14*
--
162
0.80
16.27
12.38
Modelo
Proveedor
Logo de Alibaba
QwQ-32B
24k
Abierto
13*
--
31
2.16
99.82
81.34
Modelo
Proveedor
Logo de Meta
Llama 4 Scout
131k
Abierto
10
$0.01
40
0.78
13.41
--
Modelo
Proveedor
Logo de Meta
Llama 3.3 70B
24k
Abierto
9
$0.04
50
3.58
13.56
--
Modelo
Proveedor
Logo de Meta
Llama 3.1 8B
30k
Abierto
8
--
188
0.76
3.42
--
Modelo
Proveedor

Definiciones clave

Maximum number of combined input & output tokens. Output tokens commonly have a significantly lower limit (varied by model).

Tokens per second received while the model is generating tokens (ie. after first chunk has been received from the API for models which support streaming).

Time to first token received, in seconds, after API request sent. For reasoning models which share reasoning tokens, this will be the first reasoning token. For models which do not support streaming, this represents time to receive the completion.

Average cost per task in the index. Costs are split by input, cache hit, cache write, reasoning, and answer token pricing where canonical token counts are available.

Price per token included in the request/message sent to the API, represented as USD per million Tokens.

Price per token for cached prompts (previously processed), typically offering a significant discount compared to regular input price, represented as USD per million tokens. The values shown here are the cache hit price; cache write and cache storage are billed separately and vary by provider — see "Cache pricing by provider" for detail.

Price per token to write prompt tokens into the cache so that later requests can hit them, represented as USD per million tokens. Some providers charge a premium over the standard input price to create a cache entry (e.g. Anthropic), while others cache automatically with no separate write fee.

Price per token generated by the model (received from the API), represented as USD per million Tokens.

Metrics are 'live' and are based on the past 72 hours of measurements, measurements are taken 8 times a day for single requests and 2 times per day for parallel requests.

Preguntas frecuentes

Preguntas comunes sobre Cloudflare

El modelo más inteligente disponible en Cloudflare es Kimi K2.6, con una puntuación de 44 en el Índice de Inteligencia.

El modelo más rápido en Cloudflare por velocidad de salida es Llama 3.1 8B, con 188.1 tokens por segundo.

El modelo con el menor tiempo hasta el primer token de respuesta en Cloudflare es Llama 3.1 8B, con 0.76s. Una menor latencia significa una respuesta inicial más rápida.

El modelo más económico en Cloudflare por precio combinado es Gemma 4 26B A4B, a $0.12 por 1M de tokens (proporción 7:2:1 de aciertos de caché/entrada/salida).

Los precios en Cloudflare varían hasta 6x entre modelos, desde $0.12 por 1M de tokens para Gemma 4 26B A4B hasta $0.70 por 1M de tokens para Kimi K2.6.

Sí, Cloudflare ofrece una API compatible con OpenAI, lo que facilita cambiar desde OpenAI o usar integraciones existentes del SDK de OpenAI.

10 de 11 modelos en Cloudflare admiten el modo JSON para salida estructurada.

Sí, los 11 modelos en Cloudflare admiten llamadas a funciones (uso de herramientas).

Sí, Cloudflare ofrece 7 modelos de razonamiento: Kimi K2.6, Gemma 4 26B A4B, gpt-oss-120b (high), gpt-oss-120b (low), gpt-oss-20b (high), gpt-oss-20b (low) y QwQ-32B. Los modelos de razonamiento usan pensamiento extendido para resolver problemas complejos antes de responder.

Sí, los 11 modelos en Cloudflare son modelos de pesos abiertos.

Sí, el rendimiento de un proveedor puede variar con el tiempo debido a cambios de infraestructura, balanceo de carga y actualizaciones. Medimos continuamente a todos los proveedores y mostramos las tendencias históricas de rendimiento en los gráficos "a lo largo del tiempo".

Al elegir un modelo en Cloudflare, considera: inteligencia (para tareas sensibles a la calidad), velocidad de salida (para tareas intensivas en rendimiento), latencia (para aplicaciones interactivas que requieren primeras respuestas rápidas), precios (para cargas de trabajo sensibles al costo) y funcionalidades como el tamaño de la ventana de contexto, el modo JSON o las llamadas a funciones.