Models / DeepSeek V4 FlashID: dashscope/deepseek-v4-flash
chat model by DeepSeek

DeepSeek V4 Flash

DeepSeek-V4-Flash is a high-efficiency Mixture-of-Experts model with 284B total and 13B activated parameters supporting up to a 1M-token context window. Optimized for low latency, it features hybrid sparse attention and dual-mode reasoning capabilities.

PlaygroundQuick Start
Context window1.05M
Max output393K
Maker

DeepSeek

Served On

Alibaba US

Specifications

Context window, ranked against other chat models

Proportional scale: 0 to 1.05M tokens. Not ranked, no published context window: glm-5.3-prime, qwen3-235b-a22b-2507, qwen3.7-max-2026-05-17, claude-opus-5-5.

Context window

1.0M

Source: huggingface.co
Max output

393K

Source: www.alibabacloud.com
Modality

chat

Vision
Function calling
System prompt

Benchmarks & Performance Rankings

Empirical intelligence benchmarks and category leaderboard placements for DeepSeek V4 Flash.

Compare in BenchmarksLeaderboard
Core Benchmark DimensionsSource: huggingface.co
MMLUKnowledge

0%

HumanEvalCoding

0%

Latency p50TTFT

—ms

p99: —ms

ThroughputStream

—tok/s

SWE-bench: 0%

Benchmarks

Mmlu Pro86.4
Source: huggingface.co2026-09-13
Gpqa Diamond88.1
Source: huggingface.co2026-09-13
Swe Bench Verified79
Source: huggingface.co2026-09-13
LMSYS Chatbot Arena (text)Rank #92 · Score 1436
Source: arena.ai · Leaderboard: 2026-09-112026-09-13

Pricing

Rates are read live

Pricing for deepseek-v4-flash is fetched from the live catalogue on load, so it is never served from a cache that could outlive a repricing.

Availability

Not enough data yet

We have not collected enough health probes for this model to publish an uptime or latency figure. A percentage from a handful of samples is fabricated precision, so none is shown until the sample count supports one.

How DeepSeek V4 Flash is billed

Billed at the provider's list price, with no per-token markup.

Exact list price
Tokens settle at the published rate of the model that served your call.
Cache discounts passed through
Provider prompt-caching discounts reach your balance with no added fee.
Cost on every response
Each response carries the x-nr-request-cost header, so you can reconcile spend per request.

More from DeepSeek

deepseek-v4-pro-0813deepseek-r1-distill-qwen-14bdeepseek-r1-distill-qwen-32bv3

Start building with deepseek-v4-flash

One API key, three steps. Choose your tool to begin.

  1. Get a key

    Create API key

  2. Set your API key

    export NROUTER_API_KEY="<YOUR_API_KEY>"

  3. Configure Claude Code

    Choose a tool, then copy its setup from the panel.

npm install -g @anthropic-ai/claude-code
Claude Code · dashscope/deepseek-v4-flash
# Point Claude Code to nRouter gateway
export ANTHROPIC_BASE_URL="https://api.nrouter.ai"
export ANTHROPIC_API_KEY="$NROUTER_API_KEY"

# Launch with dashscope/deepseek-v4-flash
claude --model dashscope/deepseek-v4-flash