DeepSeek V4 Flash
MoE workstation284B total, 13B active per token (MoE). Same FP4+FP8 hybrid-attention family as V4 Pro: Compressed Sparse Attention (CSA) + Heavily Compressed Attention (HCA) across 61 layers, manifold-constrained Hyper-Connections (mHC), Muon optimizer, 32T+ pretraining tokens.
- Context: 1M native, 384K max output; three modes (non-think / think-high / think-max).
The April 2026 release of V4 Flash. Superseded 2026-07-31 by DeepSeek V4 Flash 0731 - an iterative update with a 10-point Artificial Analysis Intelligence Index jump (40 to 50), agentic Elo 1189 to 1559, and Terminal-Bench 2.1 +17 to 79%. The 0731 is the model the deepseek-chat/reasoner API aliases now route to. Open weights under MIT.
AI-generated content marks
This model embeds text watermarks in generated text and adds C2PA provenance metadata to supported files such as .png, .jpg, and .svg. Marks can be lost through editing, screenshots, or format conversion, so their absence does not prove a file is human-made.
- 284.0B
- 1000k
- mit
- 🇨🇳 China
- Apr 2026
What people are building with DeepSeek V4 Flash
Real demos from X
Spark stacks roundup: V4 Flash on 2x DGX Sparks with vLLM TP=2, 62-83 tok/s single stream, full recipe published View on X →
Scores
Score per dollar
11467 pts per $/M input
general_score (86) divided by cheapest input price ($0.01/M). Higher is better value. See live pricing.
Related models
Guides covering DeepSeek V4 Flash
Save your hardware and every model page answers the real question: will it run on your machine, and how fast?
Join free - save your rig →Or run it in the cloud
Live per-provider pricing, throughput and uptime - refreshed about 3 hours ago via OpenRouter. Click a column to sort.
| Provider | Type | Input $/M | Output $/M | Cache $/M | Tok/s | Latency | Uptime | Value |
|---|---|---|---|---|---|---|---|---|
|
GMICloud
|
API | 0.09 | 0.18 | 0.018 | - | - | 99.99% | best uptime |
|
Novita
|
API | 0.14 | 0.28 | 0.028 | - | - | 99.99% | |
|
Relace
|
API | 0.01 | 1.28 | 0.008 | - | - | 99.96% | cheapest |
|
Baidu
|
API | 0.14 | 0.28 | 0.028 | - | - | 99.93% | |
|
Parasail
|
API | 0.14 | 0.28 | 0.070 | - | - | 99.93% | |
|
Cloudflare
|
API | 0.44 | 1.32 | 0.014 | - | - | 99.93% | |
|
DeepInfra
|
API | 0.09 | 0.18 | 0.018 | - | - | 99.89% | |
|
DigitalOcean
|
API | 0.10 | 0.20 | 0.020 | - | - | 99.89% | |
|
Venice
|
API | 0.10 | 0.19 | 0.020 | - | - | 99.83% | |
|
SiliconFlow
|
API | 0.13 | 0.28 | 0.028 | - | - | 99.77% | |
|
Alibaba
|
API | 0.13 | 0.27 | 0.027 | - | - | 99.14% | |
|
StreamLake
|
API | 0.04 | 0.08 | 0.008 | - | - | 97.36% | |
|
AtlasCloud
|
API | 0.14 | 0.28 | 0.028 | - | - | 97.36% | |
|
Mancer 2
|
API | 0.19 | 0.50 | 0.010 | - | - | 95.46% | |
|
OpenInference
|
API | 0.01 | 1.38 | 0.008 | - | - | 95.07% | |
|
Azure
risky
|
API | 0.21 | 0.56 | 0.031 | - | - | 94.03% | |
|
Wafer
avoid
|
API | 0.03 | 0.17 | 0.013 | - | - | 75.87% | |
| Sub | - | - | - | - | - | - | $10.00/mo Go ($5 first month) |
Default order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are "-". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.
Detailed API pricing page + JSON endpoint →
See who runs DeepSeek in production →
Inference cost over time
Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.