DeepSeek-V4-Flash-preview

Model IntroductionText

DeepSeek-V4-Flash: Next-generation flash inference for production APIs

DeepSeek-V4-Flash is the newest flash-tier model in the DeepSeek family, optimized for high-QPS chat, agents, and streaming workloads where latency and unit economics matter. It builds on the V4 architecture with a lean MoE routing path and aggressive kernel fusion so time-to-first-token stays low even under bursty traffic.

The model targets the same sweet spot as DeepSeek-V3-Flash—fast dialogue, tool calling, and structured JSON—while improving reasoning depth on coding, math, and multi-step instructions. A 128K token context window supports long documents and repository-scale prompts without forcing you onto a much larger model.

DeepSeek-V4-Flash supports standard OpenAI-compatible chat completions with streaming, function calling, and JSON mode. Pair it with prompt caching on repeated system prompts to further reduce cost on agent loops and RAG pipelines.

For teams migrating from V3-Flash, V4-Flash is a drop-in upgrade path: similar pricing tier, better quality on hard prompts, and the same deployment footprint on a single high-end GPU.

Benchmarks

Benchmark (Area)Score*
AIME 2024 (Math)96.7% accuracy (Max)
LiveCodeBench (Coding)91.6 Pass@1 (Max)
Codeforces Rating (Coding)3052 rating (Max)
SWE-bench Verified (Coding)79.0% resolved (Max)
MMLU-Pro (Knowledge)86.2 EM (Max)
HMMT 2024 Feb (Math)94.8 Pass@1 (Max)
GPQA Diamond (Science)88.1 Pass@1 (Max)
MRCR 1M (Long Context)78.7 MMR (Max)
Terminal Bench 2.0 (Agent)56.9 Acc (Max)
Apex Shortlist (Agent)85.7 Pass@1 (Max)
Chinese-SimpleQA (Chinese)78.9 Pass@1 (Max)
SuperCLUE (Chinese)68.82 overall (Max)

API request example

Endpoint /v1/chat/completions. Replace <API_KEY> with a key from the console.

Standard OpenAI Chat Completions. Swap `model` for the target slug.

curl https://api.ravenrouter.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <API_KEY>" \
  -d '{
  "model": "DeepSeek-V4-Flash-preview",
  "messages": [
    {
      "role": "user",
      "content": "Hello! Introduce yourself in one sentence."
    }
  ],
  "max_tokens": 64
}'

More models: API Reference → Examples by model

DeepSeek-V4-Flash-preview API — $0.11/1M input, $0.22/1M output | RavenRouter AI