Name: DeepSeek-V4
Brand: DeepSeek

DeepSeek-V4 by DeepSeek — Product Research | aVenture

DeepSeek-V4 is the company's 2026 flagship Mixture-of-Experts model family released under the MIT license, with V4-Pro at 1.6 trillion total parameters and 49 billion active per forward pass and a lighter V4-Flash variant for cost-efficient inference. The model supports a 128k context window via the hosted API and a 1 million token window for the open-weight release.

DeepSeek-V4 is positioned for coding, reasoning, and long-context tasks, with off-peak API pricing windows and discounted cache reads aimed at high-throughput workloads. The open-weight release allows on-premises and customized inference deployments across NVIDIA and Huawei Ascend hardware.

DeepSeek-V4 - Direct Competitors

Attribute	DeepSeek-V4	Bonsai	Frontier Open-Weight Language Models	Gemma	Kimi	Qwen
Provider	DeepSeek deepseek.com$1B raised · Series A	PrismML prismml.com$16.3M raised · Seed	Reflection AI reflection.ai$2.2B raised · Series B	Google DeepMind deepmind.google$500M raised	Moonshot AI moonshot.ai$3.6B raised · Series D	Alibaba Group alibabagroup.com$8.9B raised · Public
Founded	2023	2025	2024	2010	2023	1999
Sells To	Enterprises	—	B2B, Enterprise, Government	Enterprises	Enterprises	Consumers, Enterprise, Enterprises, Small Business
Pricing Model	Subscription	—	B2B	—	Subscription	Advertising, Subscription, Transactional
Ownership	Private	—	Private	—	Private	Public