- DeepSeek released V4 Pro (1.6 trillion parameters, 49 billion active) and V4 Flash (284 billion parameters, 13 billion active), both open-source.
- V4 is the first open model family designed from the ground up around million-token context as a default capability.
- V4 Flash is priced at $0.14 per million input tokens, roughly 50x cheaper than comparable closed-source models.
DeepSeek V4 ships two open models with million-token context
DeepSeek, the Chinese AI lab that upended Silicon Valley a year ago with its cost-efficient approach to frontier AI, released its V4 model family on Friday, April 24, 2026. The release includes two variants: V4 Pro, a 1.6-trillion-parameter mixture-of-experts model with 49 billion active parameters per forward pass, and V4 Flash, a leaner 284-billion-parameter model with 13 billion active parameters.
Both models ship with a one-million-token context window as a default, not an experimental extension. DeepSeek describes V4 as the first open model family architecturally designed around long context from the start, introducing what the company calls a Hybrid Attention Architecture that compresses attention in the token dimension rather than approximating it. The model weights are available on Hugging Face, ModelScope, and GitCode under an open license.
On benchmarks, V4 Pro leads all current open-source models in world knowledge, math, STEM, and coding tasks. According to Simon Willison's analysis, V4 "falls marginally short of GPT-5.4 and Gemini 3.1 Pro," placing the open-closed performance gap at roughly three to six months. DeepSeek has also optimized V4 for compatibility with popular agent frameworks, including Anthropic's Claude Code.
Why V4 changes the cost equation for enterprise AI
The pricing is the sharpest signal. V4 Flash at $0.14 per million input tokens and $0.28 per million output tokens undercuts every major closed-source API by an order of magnitude, according to Fortune. V4 Pro, at $1.74 per million input tokens, remains dramatically cheaper than GPT-5.5 or Gemini 3.1 Pro for comparable quality.
For enterprise teams evaluating agentic workflows that require processing entire codebases or lengthy document sets, V4's native million-token context removes a constraint that previously limited open models to short-context tasks. The release also carries geopolitical weight: according to Bloomberg, DeepSeek noted close integration with Huawei's Ascend chips, demonstrating continued progress despite U.S. export controls on advanced semiconductors.
DeepSeek V4 does not match the top closed-source models, but it no longer needs to. At 50x lower cost with million-token context and open weights, it reshapes the competitive calculus for every company choosing between proprietary APIs and running their own infrastructure.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.