DeepSeek V4 Flash Latest

DeepSeek V4 Flash Latest continues the DeepSeek family of large language models, positioned as a flash-tier variant intended to balance responsiveness with the reasoning behavior seen across recent DeepSeek releases. The model is presented through OpenRouter as a text-to-text chat endpoint that supports structured tool use and configurable sampling, which makes it a fit for assistants, agents, multi-step reasoning pipelines, and retrieval-augmented workflows where a single endpoint can be reused across routing and orchestration layers.</parameter>

Independent distribution via an NVIDIA NIM self-hosted microservice, exposed as nvcr.io/nim/deepseek-ai/deepseek-v4-flash:latest, points to a deployment-ready container image rather than a downloadable public checkpoint, and the same image surfaces an OpenAI-compatible /v1/chat/completions interface on port 8000, suggesting the weights are packaged for inference on NVIDIA hardware stacks. For practitioners, the practical advantage is that V4 Flash Latest can be reached either as a managed API call through OpenRouter or as a local container behind the same request shape, which simplifies client code across hybrid and on-prem environments.</parameter>

OpenRouter~deepseek/deepseek-v4-flash-latestdeepseek

Open provider page Provider docs

Quick Info

Provider: OpenRouter
Model key: ~deepseek/deepseek-v4-flash-latest
Release date: Aug 1, 2026
Last updated: Aug 1, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.0144
Output token cost: $0.576

Limits

Output tokens: 943,718 tokens
Context window: 1,048,576 tokens

Transparent token rates

Compare DeepSeek V4 Flash Latest pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about DeepSeek V4 Flash Latest

OpenRouter

CoverageBenchmark

Opper.ai's third-party gateway page for the DeepSeek V4 Flash (0731) checkpoint — which the OpenRouter `deepseek/deepseek-v4-flash-latest` alias currently resolves to — describes it as a retrained successor to the original V4 Flash released July 31, 2026, with identical architecture (284B-total / 13B-active sparse MoE, The same listing lists input price at $0.19/1M and output at $0.51/1M as the headline figures, and enumerates 10 routings through the Opper gateway with residency, zero-data-retention, and training-posture details — including DeepInfra ($0.09/$0.18, US, ZDR), Fireworks ($0.22/$0.66, multi-region, ZDR), Inceptron ($0.13

OpenRouter

Official sourceBenchmark

OpenRouter's first-party listing for the pinned DeepSeek V4 Flash 0423 checkpoint (slug `deepseek/deepseek-v4-flash`) describes the model as an efficiency-optimized Mixture-of-Experts from DeepSeek with 284B total parameters, 13B activated parameters, a 1M-token context window, and hybrid attention for long-context pro The same page enumerates 14 provider routings on OpenRouter — including DigitalOcean, StreamLake, Baidu Qianfan, DeepInfra, GMICloud, SiliconFlow, Alibaba Cloud International, Venice, NextBit, NovitaAI, AtlasCloud, Parasail, Mancer, Phala, and Azure (US) — each with P50 latency, tokens-per-second throughput, uptime, an

OpenRouter

Official sourceBenchmark

OpenRouter's first-party listing for the DeepSeek V4 Flash Latest model (slug `deepseek/deepseek-v4-flash-latest`) documents that this entry is a rolling alias that always redirects to the latest checkpoint in the DeepSeek V4 Flash family, with the page header referencing "Latest DeepSeek V4 Flash 0731." The page provi The same listing documents OpenRouter's reasoning-token support, explaining the `reasoning` request parameter and the `reasoning_details` array in responses so developers can surface the model's step-by-step thinking and preserve it across multi-turn conversations. It also enumerates the two available endpoints — `POST