DeepSeek V4 Flash

Speed-tuned V4 for high-volume pipelines.

Try in Playground

Input / 1M

$0.22

Cached input / 1M

$0.022

Output / 1M

$0.66

Context

LLMdeepseekServerless

Benchmarks

Published evaluation scores (higher is better). Measured output speed on our fleet, Standard tier.

SWE-bench Verified

58.4%

Real GitHub issues

GPQA Diamond

64.2%

PhD-level science

MMLU Pro

76%

Knowledge & reasoning

AIME 2025

82.7%

Competition math

IFEval

87.3%

Instruction following

Output speed

385

tok/s on runaii

Ways to serve this model

ServerlessPer-token, zero cold starts. From $0.22/M input.Start serverless →Dedicated deploymentReserved GPUs from $40/hr. Full speed control, scale-to-zero, your region.Create deployment →LoRA fine-tuneThis model is not tunable yet.Start training →

Live model traffic — last 24h

Requests to runaii/deepseek-v4-flash across the fleet, from the live meter.

Requests

…

Tokens

…

Spend

…

Quickstart

OpenAI SDK (change the base URL) or the Anthropic SDK — both speak runaii.cloud natively.

curl https://api.runaii.cloud/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $RUNAII_API_KEY" \
  -d '{
    "model": "runaii/deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Say hello in Spanish"}]
  }'

live

runaiicloud

← Model library